0% found this document useful (0 votes)

66 views

Crime Prediction Using Machine Learning and Deep L

Uploaded by

dhivadhivakar131

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

66 views

Crime Prediction Using Machine Learning and Deep L

Uploaded by

dhivadhivakar131

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 21

This article has been accepted for publication in IEEE Access.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3286344

Digital Object Identifier

Crime Prediction Using Machine

Learning and Deep Learning: A
Systematic Review and Future
Directions
VARUN MANDALAPU*1 , LAVANYA ELLURI*2 , (Member, IEEE), PIYUSH VYAS*2 , NIRMALYA
ROY1
1
Department of Information Systems, University of Maryland Baltimore County, Baltimore, MD 21250, USA.
2
Subhani Department of Information Systems, Texas A&M University - Central Texas, Killeen, TX 76549, USA.
Corresponding authors: Varun Mandalapu ([email protected]) and Lavanya Elluri ([email protected])
*These authors contributed equally to this work

ABSTRACT Predicting crime using machine learning and deep learning techniques has gained consid-
erable attention from researchers in recent years, focusing on identifying patterns and trends in crime
occurrences. This review paper examines over 150 articles to explore the various machine learning and
deep learning algorithms applied to predict crime. The study provides access to the datasets used for crime
prediction by researchers and analyzes prominent approaches applied in machine learning and deep learning
algorithms to predict crime, offering insights into different trends and factors related to criminal activities.
Additionally, the paper highlights potential gaps and future directions that can enhance the accuracy of crime
prediction. Finally, the comprehensive overview of research discussed in this paper on crime prediction
using machine learning and deep learning approaches serves as a valuable reference for researchers in this
field. By gaining a deeper understanding of crime prediction techniques, law enforcement agencies can
develop strategies to prevent and respond to criminal activities more effectively.

INDEX TERMS crime prediction, crime detection, crime datasets, deep learning, machine learning, smart
policing, survey.

I. INTRODUCTION uses artificial neural networks with multiple layers to model

Crime prediction is a complex problem requiring advanced complex relationships between inputs and outputs [4]. Both
analytical tools to effectively address the gaps in existing de- machine learning and deep learning methodologies have the
tection mechanisms. With the increasing availability of crime potential to be applied to the problem of crime prediction in
data and through the advancement of existing technology, many ways [5].
researchers were provided with a unique opportunity to study Machine learning algorithms have been utilized in crime
and research crime detection using machine learning and prediction to analyze crime data and predict future crime
deep learning methodologies. Based on the recent advances patterns [6]. For example, algorithms like decision trees, ran-
in this field [1][2][3], this article will explore current trends in dom forests, and support vector machines have been trained
machine learning and deep learning for crime prediction and on crime data from specific cities to predict crime patterns
discuss how these cutting-edge technologies are being used accurately [7]. Apart from predicting crime patterns, these al-
to detect criminal activities, predict crime patterns, and pre- gorithms can provide valuable insights into crime trends and
vent crime. Our primary goal is to provide a comprehensive patterns. These capabilities allow for deploying resources and
overview of recent advancements in this field and contribute tactics to combat crime effectively. Additionally, machine
to future research efforts. learning algorithms can also be used to identify correlations
The field of machine learning is a subset of artificial intel- between crime incidents and various environmental and de-
ligence that uses statistical models and algorithms to analyze mographic factors such as location, weather, and time of day
and make predictions based on data. On the other hand, [8]. This information can be used to develop crime prediction
deep learning methods are a subset of machine learning that and prevention strategies suitable to a given community’s

VOLUME 4, 2023 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3286344

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

specific needs. provides a comprehensive overview of recent trends in this

Predictive policing is also a significant application of ma- field and offers insights into the potential applications of
chine learning for crime prediction [9]. Predictive policing machine learning and deep learning for crime prediction.
refers to using data and analytics to inform law enforcement By highlighting the potential of these technologies and the
efforts and reduce crime. Machine learning algorithms can be challenges that must be addressed, this research article con-
used to analyze crime data from a specific geographic area, tributes to the broader research community. It advances our
such as a city or neighborhood, to identify crime hotspots understanding of the role of machine learning and deep
and predict future crime incidents. This information can then learning in crime prediction. Hence, the key contributions
be used to direct policing resources to areas where they are of this work are as follows:- first, this paper provides the
most needed, increasing the effectiveness of law enforcement amalgamation of existing studies that utilized state-of-the-
efforts. art machine learning and deep learning-based approaches in
Deep learning algorithms, such as convolution and re- the realm of detecting neighborhood crime. Thereby extend-
current neural networks, have also shown promise in crime ing the fathomable literature knowledge base. Second, this
prediction. These algorithms have been trained on crime data paper eliminates the limitation of the scarcity of potential
with either a spatial or temporal component to accurately datasets availability. We have highlighted distinct publicly
predict crime patterns in specific cities. For example, deep available datasets related to neighborhood crime prediction
learning algorithms have been used to analyze crime data, that existing studies have utilized. Thereby archiving the data
including the time, location, and type of crime incidents [10]. resources for future scholars. Third, this work drafted future
This information is used to create a predictive model that can research directions to eliminate the existing research gaps in
be used to identify potential crime hotspots and predict future neighborhood crimes. Thereby reasonably providing future
crime incidents. research objectives/questions to the research community to
Another application of deep learning in crime predic- pursue further.
tion is computer vision and video analysis. This technology
has been used to analyze video footage from surveillance II. RESEARCH METHODOLOGY
cameras to detect and classify criminal activities, such as The primary research aims to find various efficient algorithms
vandalism, theft, and assault [1]. The advanced deep learning for predicting neighborhood crimes. In our previous work [8],
models are also integrated with drones and other aerial tech- we used statistical analysis to predict the crimes in Newyork
nologies to provide new opportunities to monitor and respond city. Our paper got good attention from the researchers, so we
to criminal activities. These algorithms have also been used wanted to look for the efficient machine learning and deep
to analyze crime data from multiple sources, including crime learning approaches used in this area. We have followed a
reports, social media, and police records, providing a more systematic approach to select the papers for this review. As
comprehensive view of criminal activities [11]. By automat- part of this research, we have considered the papers from
ing this process, deep learning algorithms have the potential multiple databases related to predicting crime.
to enhance the ability to identify and respond to crime in real- For this review, we have considered all the primarily used
time, providing a crucial tool in the fight against criminal terms in the papers focused on predicting crimes. To include
activity. all the possible alternative words of each term, we have used
Despite the promise of machine learning and deep learning “*” as a wild character for IEEE and ACM databases so
for crime prediction, several challenges must be addressed. that it contains zero or more characters after the string. The
One of the biggest challenges is the availability of high- main target of this review is to check for all the existing
quality crime data. Crime data can be difficult to obtain, research works to predict crime. In addition, we want to help
and the available data may need to be completed or reliable. the research community by identifying the different datasets
Additionally, collecting and using crime data is associated used to apply the algorithms. Irrelevant studies are removed
with privacy and ethical concerns. These challenges must be by applying multiple filters to our search queries. We also
addressed to fully realize the potential of machine learning selected 30 papers to be part of the main text based on
and deep learning for crime prediction. Another challenge relevance and novelty, and 20 more papers are added in the
is the interpretability of machine learning and deep learning appendix Table 7. In this survey, we have used a combination
models. These models can be challenging to understand of an automated and manual search shown in Figure 2. In
and interpret, limiting their usefulness in decision-making. the initial stages, we focused on using the automatic digital
To effectively apply these models to the problem of crime search. In the final step, we manually read the entire paper
prediction, it is vital to develop interpretable models that can to select a set of documents in machine learning and deep
provide clear explanations of their predictions. learning areas.
Moreover, the recent advancements in machine learning Firstly, we have identified the key terms to create the
and deep learning for crime prediction show great promise in queries. We then used those keywords to construct the various
addressing this complex problem [12]. However, significant research database-related queries based on respective syntax.
challenges remain, and much work is still needed to real- Below are the queries used to explore IEEE, Science Direct,
ize these technologies’ potential fully. This research article and ACM databases.
2 VOLUME 4, 2023

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

FIGURE 1. Research paper selection methodology

IEEE query: A. DATA COLLECTION

((“Document Title”: “crime*") AND (“Document Title": We focused on looking into individual research libraries
“predic*" OR “Document Title": “detec*" OR “Document rather than searching in google scholar. Because google
Title": “recogni*" OR “Document Title": “machine learn- scholar will have data from all these databases, there could
ing" OR “Document Title":“deep learning" OR “Document be duplicates. Below are the database library homepage links
Title":“clustering" OR “Document Title":“natural language where the research works were extracted using the keywords
processing")) mentioned in search queries. Initially, we searched using all
Science Direct Query: the metadata attributes available on each database. Next, we
(“crime") AND ( “prediction" OR “detection" OR “recog- applied the filters only on the full-text papers. As we noticed
nition" OR “machine learning" OR “deep learning" OR that the number of documents is still high, we applied the
“clustering" OR “natural language processing") filter on the index terms used in the article as the results will
ACM Query: be more relevant. We have more than 450 papers from all the
“query": { Fulltext:((“crime*") AND ( “predic*" OR “de- databases at this stage. Finally, in the last step, we applied the
tec*" OR “recogni*" “machine learning" OR “deep learning" filter on the document title, where the total number of papers
OR “clustering" OR “natural language processing")) } was 157.
a) Science-Direct Elsevier (https://www.sciencedirect.com)
b) ACM (DL) Digital Library (https://dl.acm.org)
c) IEEE Xplore Access Digital Library (https://ieeexplore.
ieee.org)
After applying the automated filters as shown in Figure 1,
all the authors manually divided the work and read the papers
to select the final set. For the selection, we mainly looked into
the essential elements like the focus or objective of the article,
datasets that authors have used, algorithms applied, and the
accuracy rates. The focus of this survey is not only to help
the community know the various algorithms applied but also
to let them know about the datasets they can use to apply the
novel algorithms and get the results for their research.

III. LITERATURE ANALYSIS

A. PRE-ASSESSMENT LITERATURE ANALYSIS
An analysis of collected literature data from the distinct
research databases is essential [13] to receive information
regarding the growth of an adopted research domain, scope
across the research community, and popularity among the
existing researchers. Thus, we have performed a detailed
analysis of the collected literature data. We did pre-analysis
FIGURE 2. Steps involved for typical crime detection and post-analysis, wherein pre-analysis comprises the explo-
ration of initially collected literature ( i.e., the research papers
that were collected immediately after performing our search
query), and post-analysis comprises the investigation of those

VOLUME 4, 2023 3

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

study’s data that were finally selected after applying selec- (3 citations), and workshop (o citation) articles in the neigh-
tion criteria. Although we have exhumed the two renowned borhood crime area, this may be because this area is less
research databases, IEEE and Science Direct, lately, it has popular among the researchers.
been observed that Science direct discontinued the search
result extraction. Hence following analytical charts are based
on IEEE databases findings. Figure 3 shows the research

FIGURE 5. Distribution of article’s citations

FIGURE 3. Research publication trends from 2018-2022

publication trends in neighborhood crime from 2018 to 2022

(i.e., the last five years). It showed upward trends from 2018
to 2021 and downward trends during 2022. The years 2020
and 2021 are the apexes of COVID-19, which could be why
existing researchers have utilized that time to explore more
neighborhood crime research.
Figure 4 indicates the distribution of article page counts
for the neighborhood crime research. The majority of articles
(i.e., 46) comprise five pages. This may be because many
researchers have published their work in conferences and
symposiums rather than journals. Figure 4 depicts that very FIGURE 6. Distribution of research articles at various venues
few articles comprise more than ten pages. Since confer-
ences, seminars, workshops, and symposiums allowed the Figure 6, shows the distribution of neighborhood crime-
presentation of abstracts and short papers, this Figure 4 also related articles published in various venues. Wherein this
depicts the 1,2,3 page-long articles. figure depicts the venue name (e.g., conference, journal,
etcetera), number of papers (i..e, in numeric value-1,7..), and
percentage of published articles (i.e., 1%, 82%,..). Here, the
majority of papers have been published in the conferences
than the other venues like journals, workshops, symposiums,
and seminars. In the neighborhood crime area, 82% of the
articles were published in conferences, followed by 9% in
the journal, 4% in the symposium, 3% in book chapters,1%
in seminars, and 1% in workshops.

B. POST-ASSESSMENT LITERATURE ANALYSIS

To enhance the understanding of the neighborhood crime
domain in combination with our above-cited selection cri-
teria, we have also performed the post-assessment literature
FIGURE 4. Distribution of article’s page counts analysis because this is the crux of our literature survey to
fulfill our identified objectives. We have created a word cloud
To assess the popularity of the neighborhood crime re- for Neighborhood crime-related papers to fathom further the
search articles, we have also performed the citation analysis selected papers’ underlying key concepts or themes. A word
shown in Figure 5. Wherein the conference papers have cloud often called a tag cloud, is a graphic depiction of the
gained more citations (60) than the other journals (10 cita- terms that appear the most frequently in a given text. Each
tions), chapter (1 citation), seminar (0 citations), symposium word’s magnitude in the word cloud reflects how frequently
4 VOLUME 4, 2023

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

it appears in the text. Word clouds are frequently employed in and 2% have used the combination of ML and NLP. It has
literature reviews to swiftly pinpoint the key themes or topics been observed that machine learning techniques are popular
within a sizable body of material. Additionally, they can be in the neighborhood crime area.
used to compare various texts and find trends and patterns in Figure 9 shows the distribution of neighborhood crime-
the data [14] related selected article’s technique classes among Science
Direct and IEEE databases. As shown in Figure 9, among the
neighborhood crime-related articles, the classification task is
the prime focus utilizing various ML and DL-related tech-
niques. In the neighborhood crime domain – 63% of articles
have focused on the classification task, 29% have focused
on the regression tasks, 6% have focused on the clustering
task, and 2% of the studies have utilized the combination
of classification and clustering. It has been observed that
the Classification task is the prime focus of the studies on
neighborhood crime.

FIGURE 7. Word cloud on selected articles

FIGURE 9. Distribution of neighborhood crime-related selected article’s

technique’s classes

Figure 10 shows the Distribution of neighborhood Crime

related selected article’s Technique classes Versus technique
FIGURE 8. Distribution of neighborhood crime-related selected article’s
Type among Science Direct and IEEE databases. Figure 10
technique types answers the question- what AI Technique is used for which
technique classes (classification, clustering, and regression)?
Figure 7 shows the word cloud for the neighborhood Herein all five techniques (ML, DL, DL+NLP, ML+NLP, and
crime-related selected articles wherein many crime-related ML+DL) have been used for the classification task whereas,
key terms have emerged as trends in the existing studies. for clustering and the combination of clustering and classifi-
“crime”, “criminal”, “policing”, “enforcement”, and “secu- cation, the ML is solely used. In neighborhood crime articles
rity” are the words that indicate the emphasis of researchers NLP is also used for classification and regression tasks in
on the sub-areas of crime detection. Moreover, “prediction”, addition to ML and DL. For the classification and regression
"algorithms", and “techniques” are the words that indicate the tasks ML and DL both have been used.
aim of the existing studies. This word cloud is aligned with
our selection criteria and objectives that further validate our IV. CRIME PREDICTION PROCESS & DATASETS
final set of selected articles for this literature review study. Crime prediction using machine and deep learning involves
Figure 8 shows the Distribution of neighborhood crime- several major steps as shown in Figure 11. The first step
related selected article’s technique types among Science Di- is data collection, which involves gathering relevant data
rect and IEEE databases, respectively. As shown in the Figure such as crime statistics, demographics, and weather patterns.
8, among the neighborhood crime-related articles, machine The next step is data preprocessing, which includes cleaning
learning (ML), the combination of machine learning and deep and transforming the data into a usable format. After data
learning (DL), and the combination of machine learning and preprocessing, the data is split into training and testing sets
natural language processing (NLP), the combination of DL for model development and evaluation. The next step is
and NLP, and DL are the majorly used AI technique type. In feature engineering, which involves selecting relevant fea-
the neighborhood crime domain – 67% of articles have used tures from the data that can be used to train the model.
ML, 21% have used DL, 8% have used the combination of Once the features are selected, various machine and deep
ML and DL, 2% have used the combination of DL and NLP, learning algorithms can be applied to the data for training and
VOLUME 4, 2023 5

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

FIGURE 10. Distribution of neighborhood Crime related selected article’s Technique classes Versus technique Type

prediction purposes. Finally, the trained models are evaluated V. CRIME PREDICTION USING MACHINE LEARNING
using various performance metrics to assess their accuracy TECHNIQUES
and effectiveness in predicting crime. The results can be used Traditional machine learning models have proven to be ef-
to support decision-making in law enforcement and crime fective for crime prediction. Various types of models such as
prevention efforts. decision trees, support vector machines, logistic regression,
As shown in Table 1, there have been many datasets and random forests have been utilized to analyze crime data
used in crime detection and prediction research articles. One and identify patterns that can be used to predict criminal
example is the Chicago Crime Dataset, which contains data activity. Unlike deep learning, which relies on large amounts
on crimes reported in the Chicago area. This dataset has of data and complex neural networks, traditional machine
been used to create models that predict the likelihood of learning models require fewer data points and are easier to
specific types of crimes occurring in different areas of the interpret. For example, a logistic regression model can be
city. Another dataset used in crime prediction research is used to predict the likelihood of a certain type of crime
the London Crime Dataset, which contains data on crimes occurring based on factors such as time of day, location,
reported in London city. This dataset has been used to create and demographics of the area. A decision tree model can
models that predict the likelihood of crimes occurring in be used to identify the most important factors that contribute
specific areas and their relationship to the socio-economic to the occurrence of a particular crime. Random Forest (RF)
factors of people based on their geo-locations in the area. models can be used to analyze a wide range of features and
Other datasets commonly used in crime detection and pre- make predictions about crime patterns. In addition to these
diction research include the Los Angeles Crime Dataset, the techniques, traditional machine learning models can also be
New York City (NYC) Crime Dataset, and the Philadelphia used for anomaly detection and outlier analysis in crime data.
Crime Dataset. These datasets contain information on crimes By identifying unusual patterns or outliers in the data, law
reported in their respective cities and have been used to create enforcement agencies can detect potential criminal activity
models that predict the likelihood of specific types of crimes and take action to prevent it. In the below sections 5.0.1
occurring in different areas. In addition to these, there are and 5.0.2, we discuss the latest research on using machine
also global datasets that focus on CCTV video footage, types learning model-based regression and classification for crime
of aggression, and weapons for real-time crime predictions. prediction.
Overall, these datasets provide valuable information for
researchers to build crime prediction models that could help 1) Machine Learning based Regression Methods for Crime
law enforcement agencies prevent and respond to crimi- Prediction
nal activities more effectively. The location and access to Several crime detection scenarios are predicted using regres-
datasets used by research articles surveyed in this paper are sion techniques as shown in 2. Researchers mainly focused
listed in Table 1. on prevalent crimes like motorcycle robbery, losing property,
and crimes in urban areas. Numerous factors may drive
the boom in motorcycle robberies. For example, population
growth and density, commuting conduct, bike usage, etc.
6 VOLUME 4, 2023

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

FIGURE 11. Architecture flow of crime prediction

These situations are problems for the police to govern and ous predictive solutions use intelligent systems to identify
screen regularly since it requires forecasting and probabilities when will a criminal offense will arise, which lets police send
of robbery in a precise term. A novel method is proposed in to those areas that are in danger. As part of their research,
the research study [49]; the authors created an application the authors looked into four machine-learning approaches for
to predict motorcycle robbery with a technique to consider identifying where a criminal offense will arise in Fortaleza,
outside consequences using ARIMAX – TFM with a single Brazil. Their results indicate that easy algorithms are efficient
input. The accuracy of ARIMAX is measured using Mean in predicting crime. Also, they have seen that the Decision
Absolute Percentage Error (MAPE) and Root Mean Squared Tree and Bagging Regressor strategies obtained quality pre-
Error (RMSE), and the scores are 32.30 and 6.68. diction outcomes.
Rapid urbanization is a compelling challenge connected As mentioned above, numerous linear models are there to
to city management and services. Cites with higher crime predict crime through the correlations between urban metrics
rates are difficult to manage public safety. To reduce crimes, and crime. However, due to multicollinearity and nonGaus-
new technologies are relieving police departments to access sian distributions in urban attributes, we usually tend to make
vast amounts of crime data to identify underlying trends controversial conclusions on these attributes to predict crime.
and patterns. These technologies have doubtlessly grown the Ensemble-based machine learning algorithms can deal with
efficient deployment of police assets within a given region such problems adequately. In the research work [44], authors
and ultimately guide greater powerful crime prevention. Re- applied random forest regressor to predict the crime and
searchers have worked on predictive models to use these quantify the impact of urban attributes on homicides. Their
datasets and predict crimes. Study [23], provides a technique approach has 97% accuracy in crime prediction, and the
primarily based on spatial analysis and auto-regressive fash- significance of city indicators is clustered and ranked equally.
ions to automatically locate excessive-hazard crime areas in Their research identifies the rank of urban indicators based
city areas and reliably forecast crime tendencies in each area. on their significance in predicting crime. As per their results,
Experiments are performed on real-world datasets gathered unemployment and illiteracy are the essential variables for
in New York City and Chicago. depicting homicides in Brazilian towns.
Another study [50], compared multiple techniques to pre-
dict the crimes in various areas of a metropolis. This research 2) Machine Learning based Classification Methods for Crime
explored three predictive models: linear regression, logistic Prediction
regression, and gradient boosting. The authors utilized fea- Traditional regression techniques can successfully check the
ture selection techniques to select essential predictors. By variables’ significance but, they must be more reliable for
applying feature selection methods, there is an improvement crime prediction. In many research works [32], [33], [34]
in accuracy scored, and it helped to avoid model overfitting. mentioned in section V-2, authors have proven that machine
After comparing the results of all four models, the authors learning models effectively predict crimes. Still, they could
found that the gradient boosting technique outperformed, be more efficient in identifying which variables are signifi-
proven to be the best method to predict the crime rate in the cant in predicting crimes. We further examined the classifi-
urban area. cation techniques to predict different criminal incidents like
In another study [51], authors have looked at the crimes in analyzing the criminal reports as shown in Table 3. Studying
Brazil, which have increased rapidly in recent times. Numer- those reviews for crime prediction enables regulatory author-
VOLUME 4, 2023 7

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

TABLE 1. Crime detection papers with data science algorithms

Dataset Links Type of Data Country - Locality

https://data.gov.in/resource/incidence-crime-committed-against-women-india-
Crime India
during-2001-2012 [15]
India - Madhya
https://indore.mppolice.gov.in/ [16] Crime
Pradesh
https://ncrb.gov.in/en/node/3721 [17], [3], [16], [18], [19] Crime India
https://data.gov.in/catalogs [17], [20] Crime India
https://www.kaggle.com/datasets /murderaccountability/homicide-reports [21] Crime Unites States
https://geodash.vpd.ca/opendata/ [6] Crime Canada - Vancouver
Crime - Spatio Unites States -
https://plenar.io/explore/ [22], [23]
Temporal Chicago
Unites States -
https://data.cityofchicago.org/ [24], [25], [26], [27], [28], [29], [30] Crime
Chicago
United States - San
https://datasf.org/opendata/[31], [28] Crime
Francisco
United Kingdom -
https://data.london.gov.uk/dataset/ [32] Crime
London
https://br-city.survey.okfn.org/place/rn.html [33] Crime Brazil - Natal
United States -
https://opendata.cityofnewyork.us/data/ [30], [34], [27], [29], [35], [23], [8] Crime
Newyork
www.twitter.com [36], [37], [38] Social Media Global - Twitter
United States - San
https://www.kaggle.com/competitions /sf-crime/data [19] Crime
Francisco
United States - Los
https://data.lacity.org/ [39], [29] Crime
Angeles
https://dasci.es/transferencia/open-data/24705/ [40] Crime- Weapons Global
https://www.gdeltproject.org [41] Crime Global
www.foursquare.com [29] Geospatial Unites States
United States -
https://www.nyc.gov/taxi [29] Taxi
Newyork
https://www.wunderground.com/weather /api [29] Weather Global
https://www.icpsr.umich.edu/web /NACJD/studies/3355/datadocumentation [42] Crime United States
https://roc-ng.github.io/XD-Violence/ [43] Crime - Video Global
www.facebook.com [38] Social Media Global - Facebook
https://datasus.saude.gov.br/ [44] Health Brazil
Crime - Spatio Unites States -
https://www.phillypolice.com/crime-maps-stats/ [45]
Temporal Philadelphia
https://mapstyle.withgoogle.com/ [45] Geospatial Global
https://www.mapbox.com/ [45] Geospatial Global
https://data.police.uk/data/ [46] Crime United Kingdom
https://polisen.se/en/services-and-permits/police-record-extracts/ [47] Crime Sweden
https://www.openu.ac.il/home/hassner /data/violentflows/ [48] Crime - Video Global

ities to deal with crime prevention strategies. However, col- Another research [53], focuses on predicting crime using
lecting these reviews personally and determining their crime the XGBoost algorithm. Based on the records of theft in-
types is challenging. In one study [52] authors have created a stances in H city, they developed an optimized decomposition
novel approach, an incremental classifier that learns the new and fusion method based on XGBoost and applied multi-
data and dynamically predicts the results. In this research class classification models like OVR-XGBoost and OVO-
[52], they have utilized the Bi-objective Particle Swarm XGBoost. As the theft datasets have different classes, they
Optimization technique to develop an efficient incremental have utilized the SMOTENN algorithm to process and make
classifier for dynamically classifying and predicting crime data the dataset balanced. Their results show that OVR-
reports. Crime reports from various countries have been col- XGBoost and OVO-XGBoost models’ prediction accuracy is
lected from online newspapers to measure the performance better than the baseline XGBoost models. In the study [54],
of their classifier. Also, they evaluated the results manually the authors have selected 17 variables for crime prediction,
with unprejudiced police witness narrative crime reports. and the XGBoost algorithm is adopted to train the prediction
They tested their approach on four datasets to measure their model. A post hoc interpretable approach, Shapley additive
model’s statistical significance. explanation (SHAP), is used to parent the contribution of
8 VOLUME 4, 2023

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

TABLE 2. Crime prediction using machine learning regression techniques

Methodology ML Algorithms Dataset Performance

ARIMAX method is used to predict the time series
data from theft cases that affected the number of ARIMAX algorithm City of Yogyakarta RMSE - 6.68
motorcycles. [49]
Chicago Crimes -
2016 RMSE for Area -
Crime data is used to detect and map crime-dense 2001 to
ARIMA, Random
143.73, CDR 1 - 57.8,
region (CDR) in NYC and Chicago, then ARIMA is present, New York
Forest (RF), RepTree
CDR2 - 29.85 & CDR 3 -
applied on these crime hotspots. [23] City Crimes:, 2006
and ZeroR
16.19
to 2016
Clustering is used to understand distribution across city Linear regression (LR), Saint-Petersburg
and then cluster are used to predict the number of LoR and Gradient Russia Crime R-Square - 0.9
crimes per location. [50] boosting. (2014 - 2017)
Four machine learning methods are applied to predict
RF, Bagging, DT, Extra City of Fortaleza
the location of where a crime, more specifically, where RMSE - 0.00231
tree data
a crime of robbery will occur. [51]
Department of
Random forest regressor is used to predict crime and
Informatics of the Upto 97% Accuracy,
quantify the influence of urban. Results showed that Random Forest
Brazilian Public Adjusted R square 80% on
unemployment and literacy are the two major Regressor
Health System — average.
indicators of crimes. [44]
DATASUS.

person variables. SHAP, a post hoc interpretable method, is semble classification approach for predicting crime reports
used to determine the significance of individual variables. better than traditional classifiers. Crime reports are graphi-
Among all 17 variables used in this research, the percentage cally modeled to locate the maximal independent subset of
of the non-neighborhood population and the populace aged features, and then they use decision tree classifiers on this
25–44 contribute greater than different variables in predicting set. Extensive experiments are performed to compare the
crime. The higher the ambient population of elderly 25–44 in overall performance of the proposed approach on numer-
the vicinity, the more public crimes. The authors have also ous crime data sets. The developed ensemble classification
validated the SHAP values to demonstrate each variable’s model demonstrated better performance. Apart from predict-
contribution to the crime prediction across the experimental ing crime, researchers [42], [43] also focused on interpreting
findings. These outcomes of the neighborhood techniques crime-related predictions. This will lead to a better under-
can assist the police in identifying the most important fac- standing of what impacts crime detection.
tors.At the same time, the global model identifies the essen-
tial features of the entire region. A. CRIME PREDICTION USING DEEP LEARNING
Another research [55] focuses on predicting crime during TECHNIQUES
or after psychiatric care. As modern threat-evaluation equip- Deep learning has become a popular method for crime pre-
ment is time-consuming to administer and offers constrained diction in recent years. The studies included in the reference
accuracy, this research looked to expand a predictive model research articles use a range of deep learning algorithms,
designed to discover psychiatric patients liable to commit such as convolution neural networks (CNN), deep neural
the crime. The authors utilized the longitudinal nice of the networks, and sentiment analysis, to analyze various types of
affected Danish person registries, recognizing the 45.720 data, including text, images, audio, and social media. These
adult patients who had connected with the psychiatric system algorithms are capable of detecting patterns and anomalies
in 2014, of which 474 committed crimes leading to a forensic in the data that can indicate criminal activity. One of the
psychiatric treatment direction after discharge. Authors have key strengths of deep learning is its ability to handle large
used four gadget studying models (Random Forest, Logistic and complex datasets, making it well-suited to the task of
Regression, XGBoost, and LightGBM) over various sociode- crime prediction. For example, image analysis algorithms can
mographic, judicial, and psychiatric variables. Their model detect threatening objects in crime scenes and predict the
identified 47% of future forensic psychiatric patients, making likelihood of a crime occurring. Text mining techniques can
correct predictions in 57% of samples. This research demon- be used to analyze crime-related tweets and make predictions
strates how a clinically useful preliminary risk assessment about crime patterns. In addition, deep learning algorithms
is achieved using machine learning classification techniques. can detect anomalies in crime data in smart cities, which
Their research helps to flag possible forensic psychiatric could indicate the presence of criminal activity. Researchers
patients while in contact with the general psychiatric system, used these techniques to tackle both regression and classifi-
which allows early intervention initiatives to be activated. cation problems in crime prediction as detailed in the below
Another research work [56] presents a graph-based en- sections.

VOLUME 4, 2023 9

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

TABLE 3. Crime prediction using machine learning classification techniques

Methodology ML Algorithms Dataset Performance

OVR-XGBoost: One Vs Rest OVO-XGBoost: One Vs One.
XGBoost based H City China:
The primary difference is based on how the dataset is Aggregate Accuracy - 85%
algorithms (2019)
organized for classification. [53]
USA, India & UAE
Applying particle swarm optimization-based classifier and a Multiple classification
crime articles Aggregate Accuracy - 79%
rules engine to classify crime reports. [52] algorithms
(2007 - 2017)
17 spatiotemporal variables were used to train and test the ZG City, in
Accuracy - 89% & AUC -
XGBoost algorithm, and then SHAP is used to explain the XGboost model Southeast China.
0.586
model predictions. [54] (2017 to 2020 ).
Danish national
Logistic Regression
Random under and over sampling techniques are used to psychiatric patient
(LoR), RF, XGBoost F1 Score - 76%
evaluate multiple algorithms. [55] register, (474
and LightGBM
crime).
USA, UAE & India
Graph theory based. Decision trees and heuristics are used for Multiple classification
Crime Articles Aggregate F1 Score - 88%
feature selection and an ensemble classifier is applied. [56] algorithms
(2008 - 2016)
Identify if the crime is spread uniformly according to
London crime
population density or whether specific socio-economic Visual Analysis Aggregate Accuracy - 97%
dataset
attributes account for increased or decreased crime. [32]
Follows an algorithm-as-a-service architecture (AaaS),
K Nearest Neighbor Crime endogenous
providing insights into existing public safety systems and Aggregate Accuracy - 90%
(KNN) data sources
platforms. [33]
Prediction of crime in neighborhoods of New York city using XGBOOST, RF and New York crime
Aggregate Accuracy - 52%
spatial sata analysis. [34] SVM data
EADT approach is used for Interpretable and Accurate Crime Pennsylvania state Aggregate Accuracy -
Decision Tree (DT)
Prediction. [42] prisons 77.6%
National Crime
Traditional data science steps are used to cluster crimes and
K-Means Records Bureau Aggregate Accuracy - 78%
then predict using classification modeling methods. [18]
(India)

TABLE 4. Crime prediction using deep learning regression techniques

Perfor-
Methodology DL Algorithm Dataset
mance
Long Short Term Memory
A Graph Convolution network in combination with ST-ResNet (LSTM) in combination
is used to perform spatiotemporal analysis and then LSTM is with Spatio Temporal Chicago Crime Data R-Squared
used to detect crimes in each community. GBDT is used to Graph Convolution (2001 - 2020) - 0.84
combine outputs of GCN and LSTM. [26] Network (ST-GCN) and
Gradient Boosting Tree
Developed Attention-LSTM to process categorical-temporal
ATTN-LSTM, San Francisco and Aggregated
data and Stacked Bidirectional LSTM model to process spatial
St-Bi-LSTM, Fusion Chicago Crime Data R-Squared
information. The two were fused using feature and
Models (2004 to 2017) > 0.90
decision-level fusion. [28]
LA Crime data is separated into block-wise information based Mixed Spatio Temporal
Los Angeles Crime RMSE -
on the hour of the day, area, and city. These blocks are used to Neural Network based on
Data 0.22
train CNN model. [39] CNN
New York City,
A model framework based on three phases i.e., intercity
CNN based on Chicago, and Los
similar-grid matching, auxiliary features construction, and RMSE -
unsupervised domain Angeles Crime,
crime risk prediction using a dense CNN-based unsupervised 0.62
adaptation model (UDAC) weather and taxi Data
domain adaptation. [29]
(2015)
Xiaogan, data (2016). Accuracy
Weather, holiday, time slot ID, and Day of the week for spatial New York City data Xiaogan
LoR, Support Vector
dependency and GRU-based features for Temporal (2014). Additionally Dataset -
Regressor, RF, GRU, Deep
dependencies to predict a number of incidents in different POI, Urban anomaly 54%, NYC
Crime and T-GCN
locations. [57] and Weather data are Dataset -
used. 61%

10 VOLUME 4, 2023

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

1) Deep Learning based Regression Methods for Crime holiday, time slot ID, and Day of week information to extract
Prediction spatial dependency (distance graph, poi similarity, and crime
Deep learning algorithms in regression analysis are used as a similarity). Temporal dependencies captured using GRU are
tool for crime prediction to identify the factors most strongly used to predict the number of incidents in different locations.
associated with crime and use these relationships to make These research articles highlight the versatility of regres-
predictions about future crime patterns. The research articles sion as a tool that can be integrated with other techniques
in this area highlight the strengths of regression in modeling to enhance the performance of crime prediction models.
the relationship between multiple variables, including crime Another commonality among the papers is the use of re-
data, weather data, demographic data, social media data, gression to model the relationship between crime data and
and location data. A common theme among these research other variables, such as weather and demographic data, to
articles shown in Table 4 are the use of regression combined incorporate external information that may influence crime
with deep learning techniques, such as convolution neural patterns. This allows for the creation of more comprehensive
networks, recurrent neural networks, attention mechanisms, and accurate models of crime patterns. In summary, the five
and sequential fusion models, to improve the accuracy of research papers demonstrate the strengths of regression as
crime prediction. a tool for crime prediction, including its ability to model
In research focusing on theft crime prediction [26], the au- the relationship between multiple variables, its versatility
thors use regression to model the relationship between theft in being integrated with other techniques, and its ability to
crime data, demographic data, and weather data. The regres- incorporate external information that may influence crime
sion model adopts two deep learning models, a Long Short- patterns.
Term Memory (LSTM) network and a Spatio-Temporal
Graph Convolutional Network (ST-GCN), to predict the like- 2) Deep Learning based Classification Methods for Crime
lihood of theft crimes in urban communities. The regression Prediction
model can incorporate external information, such as weather Deep learning algorithms are trained on large amounts of data
data, which can influence crime patterns. The LSTM and ST- to classify instances into various categories. This makes them
GCN models capture the temporal and spatial dependencies ideal for solving classification problems in crime detection.
in the data, respectively. In another article [28], the authors Deep learning models can accurately organize criminal ac-
use regression to model the relationship between crime data, tivity and detect criminal intent by analyzing vast amounts of
weather data, and social media data. The regression model is data, including images, audio, text, and social media data. For
part of a more comprehensive, multi-module approach that example, image-based data can provide detailed information
uses attention mechanisms and sequential fusion models to about crime scenes, including the presence of weapons and
predict the likelihood of crimes. This framework consists of other objects that may indicate criminal intent. Similarly,
four sub-modules, where the initial two modules adopt St- audio-based data can provide valuable insights into the tone
BiLSTM and ATTN-LSTM to process temporal and spatial and context of a conversation, helping to identify potential
features. Finally, two fusion models are used to abstract illegal activities. Another advantage of deep learning for
the data and make crime predictions on Chicago and San classification problems in crime detection is the ability to
Francisco crime datasets. identify hidden patterns in the data that traditional meth-
In another research focused on using spatiotemporal data ods may miss. For example, deep neural networks can be
[39], the authors use convolutional neural networks to de- trained to analyze crime-related tweets, uncovering patterns
velop a regression model on publicly available crime data that indicate a potentially criminal act. The results of deep
in Los Angeles. The regression model is part of a more learning models in crime detection have been awe-inspiring.
extensive, mixed spatiotemporal neural network that is de- The papers reviewed under deep learning classification are
signed to make real-time predictions about the likelihood of listed in Table 5
crimes. The authors claim that using regression in combina- In crime-related classification, two main types of deep
tion with the diverse spatiotemporal neural network results learning algorithms are used: Convolutional Neural Net-
in improved accuracy and real-time performance. Another works (CNN) and Recurrent Neural Networks (RNN).
research [29] that applies crime risk prediction across differ- CNN’s are commonly used in image-based classification
ent cities uses regression to model the relationship between tasks, including crime scene prediction. In the research article
crime data and demographic data from other cities. The re- focusing on crime scene data [59], CNNs are trained to
gression model is part of an unsupervised domain adaptation detect threatening objects in crime scenes, such as weapons.
technique designed to predict the likelihood of crimes in new This allows the model to comprehensively analyze the crime
cities. The authors claim that using regression in combination scene, including the presence of things that may indicate
with the unsupervised domain adaptation technique results criminal intent. On the other hand, RNN’s are commonly
in improved accuracy in crime prediction. A recent research used to study temporal patterns in data. In a research article
article [57] applied machine learning and deep learning focusing on crime prediction based on behavioral tracking
methods to crime data from Xiaogan, a medium-sized city [40], the authors use a combination of deep learning al-
in China, to predict crime hourly. The models use weather, gorithms, including CNN and RNN, to analyze behavioral
VOLUME 4, 2023 11

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

TABLE 5. Crime prediction using deep learning classification techniques

Methodology DL Algorithm Dataset Performance

VGGNET19 based
R-CNN algorithm is adopted detect multiple objects relevant to Youtube and Google Train Accuracy
on Fast RCNN &
crime in real time videos and images. [58] Videos & Images - 100%
RCNN
2000 Images of
CNN is used to extract features from CCTV images to predict
CNN weapons and blood Accuracy - 90.2%
crime scene objects. [59]
scenes
Classify object or person in video feed and track abnormal activity
1 Million video Mean Squared Error
using Deep CNN (DCNN) and Recurrent Neural Network (RNN). DCNN and RNN
streams [[61] (MSE) - 1.38e − 10
[60]
CNN, KNN, NB,
Extract tweet data and process text to identify text keywords that Support Vector Saudi Arabia tweets
Accuracy - 79%
are relevant to weapons used for crime or criminal activity. [36] Machine (SVM), (2017 - 2021).
DT and RF
RNN, Gradient
Live CCTV data is used to detect faces and weapons using CNN-
Recurrent Unit CAVIAR Dataset Accuracy - 95.97%
GRU model. [40]
(GRU) and LSTM.
Traditional ML
Accuracy for Crime
Preprocess crime data to extract features and train both traditional classifiers and Saudi Arabia (2018 -
Type - 99% & Crime
and Deep Learning models to predict crime region and type. [41] Artificial Neural 2020)
Region - 81%
Network (ANN)
Audio data is used to extract Mel-Frequency Cepstrum
Coefficients from sound waves and a CNN-RNN classifier is XD-Violence Audio &
CNN & BERT Accuracy - 85.63%
deployed & text data, features are extracted for BERT. Both Video Data
models are combined by fusion model. [43]
Twitter data is processed by keyword filtering and labeled. The
100000 Tweets
feature vectors are generated to be inputted into models for SVM and ANN Accuracy - 90.3%
(January 2020)
training. [38]
Deep
Crowd Violence,
Video data is used to extract spatiotemporal features and gestures Reinforcement Aggregate F1 score -
UCSD and Violent
to train DRNN for classification (Hostility & Violence). [48] Neural Network 78%
Flow Data.
(DRNN)
Deep Learning and Machine learning models are applied to predict SVM, LR, DT, RF,
2018 NYPD and
different types of crime and their relation to weather in Newyork CNN, RNN, LSTM Accuracy - 99%
Weather data
city. [8] and GRU

tracking data and motion analysis data. The study shows that multimodal deep learning model based on CNN and BERT
this approach can effectively predict criminal activities, such that considers both audio and text data to classify crime-
as theft and robbery, by analyzing patterns in the behavior related events. This model is advantageous when audio data,
and movements of individuals in a given area. In the research such as 911 calls, is available and can provide a complete
articles [36] focusing on social media data, Artificial Neural picture of the crime event.
Networks are trained on crime-related text data to predict the These research articles including multiple other studies
likelihood of a crime occurring. These models analyze the [38], [48] highlight that deep learning algorithms, including
context and tone of the text data to classify patterns that may CNN and RNN, have successfully applied to various data
indicate criminal activity. types for crime prediction and classification. These studies
In addition to the articles mentioned above, several other demonstrate the versatility of deep learning algorithms in
research studies [58], [41], [8] have also used deep learning this field and provide valuable insights into the factors con-
techniques for crime prediction and classification. These tributing to criminal activity. By leveraging the strengths of
studies demonstrate the versatility of deep learning algo- these models, law enforcement agencies can gain a more
rithms in crime-related classification tasks, as they can be comprehensive understanding of criminal activity and take
applied to a wide range of data types, including images, proactive measures to prevent crime from occurring.
text, audio, and social media data. For example, research
focusing on crime anomaly detection [60] uses deep learning VI. DISCUSSION AND FUTURE WORK
algorithms, including Autoencoders and CNN, to analyze The adoptions of machine and deep learning algorithms
crime patterns in smart cities. The study shows that this to predict or detect crime has shown great promise in ad-
approach can effectively detect unusual crime patterns, which dressing this complex problem. By utilizing vast datasets
may indicate the presence of criminal activity. In another and advanced algorithms, these technologies can potentially
study focusing on audio and text data [43], the authors use a improve the accuracy and effectiveness of crime prediction
12 VOLUME 4, 2023

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

TABLE 6. Future Research Directions

Concept Areas and Research Questions

Feature-Oriented
1. How do visual features (e.g., images/videos) affect (i.e., positively or negatively) the crime
identification?
2. How do audio features (e.g., voiceover, call recordings) affect (i.e., positively or negatively) the crime
identification?
3. How do Textual features (e.g., investigating notes, complaint statements, crime interviews) affect (i.e.,
positively or negatively) the crime identification?
4. What are the visual features to be considered to detect neighborhood crime?
5. What are the Audio features to be considered to detect neighborhood crime?

Technique-Oriented
1. Are there any reinforcement learning techniques available to detect neighborhood crime?
2. Are there any transfer learning techniques available to detect neighborhood crime?
3. Are there any generative adversarial neural network techniques available to detect neighborhood crime?

models. However, despite the advances in this field, there accurate and reliable data. In order to effectively apply ma-
are still significant gaps in the current understanding of how chine learning and deep learning for crime prediction, it is
these technologies can be effectively applied to the problem important to have access to high-quality and up-to-date crime
of crime prediction. In this section, we will discuss the data [71]. As this study showed that many earlier researchers
potential benefits of machine learning and deep learning took advantage of data related to demographics, whether
algorithms for crime prediction and the future research. outside of crime-relevant datasets, there is a need to develop
One of the primary advantages of machine learning and algorithms that can accurately handle data from multiple
deep learning algorithms for crime prediction is the ability sources and integrate it into a single predictive model.
to analyze large datasets and identify patterns in criminal Another significant area of focus should be to research
activity or behavior. The ability of these algorithms to pro- more on the ethical implications [72], [73] of using machine
cess vast amounts of data, including social media and other learning and deep learning for crime prediction. As these
online sources [62], [63], can provide valuable insights into technologies are used to predict individuals and communi-
criminal activities that are yet to be committed. Furthermore, ties, it is important to ensure that they do not perpetuate exist-
deep learning algorithms like CNN and RNNs have been ing biases or lead to discrimination [74], [75]. Furthermore,
used to analyze video footage from security cameras [64]. there is a need for more research on the privacy implications
This capability provides a more accurate and efficient means of using these technologies for crime prediction [76], [77],
of detecting criminal activities. Another major benefit of [78], this included but not limited to the potential risks of data
machine learning and deep learning for crime prediction breaches and the misuse of personal information. Another
is the ability to develop real-time prediction models [65]. significant gap in the existing research is the need for more
These models can be used to analyze crime data in real- research studies on the effectiveness of machine learning and
time and to predict future crime incidents. This supports law deep learning for crime prediction in the real world [79].
enforcement agencies to act quickly if a criminal activity While these technologies have shown great promise in this
is being committed. Additionally, integrating decentralized area, there is a need for more rigorous evaluations of their
machine learning algorithms with wearable technology, such accuracy and effectiveness [80] in real-world scenarios. Ad-
as body cameras and smartwatches [1], [66], provides new ditionally, there is a need for more research on the scalability
opportunities to collect and analyze data related to criminal of these technologies and the challenges associated with their
activities. implementation in large-scale systems.
Even though machine learning and deep learning algo- Overall, machine learning and deep learning methodolo-
rithms support effective crime prediction, there are still some gies have the potential to transform the field of crime pre-
significant challenges that needs to be addressed. One of the diction by providing more accurate and effective methods
major challenges in this area is the need for interpretable for predicting criminal activities. However, in order to fully
models [54], [67] that can provide clear explanations of realize the potential of these technologies, it is important to
their predictions. This is particularly important in the context address the existing research gaps and challenges, including
of crime prediction, as incorrect predictions might lead to the need for interpretable models, accurate and reliable data,
serious consequences for individuals and communities [68]. ethical considerations, and more rigorous evaluations of their
Apart from the existing model-based explanation methods, accuracy and effectiveness. By addressing these gaps, we can
it is also important to incorporate causal based explanations advance our understanding of the role of machine learning
[69], [70] that focus on cause and effect relationship be- and deep learning algorithms in crime prediction and con-
tween crime patterns and relevant feature variables. Another tribute to the development of more effective and efficient
challenge that needs to be addressed is the need for more policing strategies.

VOLUME 4, 2023 13

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

As a future research goal and agenda, we have illustrated a [2] Soon Ae Chun, Venkata Avinash Paturu, Shengcheng
range of prospective research directions in the area of neigh- Yuan, Rohit Pathak, Vijayalakshmi Atluri, and Nabil
borhood crime based on the importance and also current lack R. Adam. Crime prediction model using deep neural
of focus in the areas. From Table 6, future researchers may networks. In Proceedings of the 20th Annual Inter-
want to address concerns like - "Are there any reinforcement national Conference on digital government research,
learning techniques available to detect the neighborhood pages 512–514, 2019.
crime?", and What are the visual features to be considered [3] Sapna Singh Kshatri, Deepak Singh, Bhavana Narain,
to detect the neighborhood crime? The relevant datasets Surbhi Bhatia, Mohammad Tabrez Quasim, and
are available for such identified future research questions; Ganesh Ram Sinha. An empirical analysis of machine
such data may be utilized to accomplish the goal of early learning algorithms for crime prediction using stacked
recognition of neighborhood crime. generalization: An ensemble approach. IEEE Access,
The presented literature base and futuristic research goals 9:67488–67500, 2021.
offer a number of elements to direct future studies and [4] Christian Janiesch, Patrick Zschech, and Kai Heinrich.
thus theoretically support the effort to identify neighborhood Machine learning and deep learning. Electronic Mar-
crimes. Our systematic review offers a thorough grasp of the kets, 31(3):685–695, 2021.
characteristics and methods used by earlier research to recog- [5] Wajiha Safat, Sohail Asghar, and Saira Andleeb Gillani.
nize and detect crimes. In addition, we have outlined 8 more Empirical analysis for crime prediction and forecasting
research questions that fall into the categories of technique- using machine learning and deep learning techniques.
oriented questions and feature-oriented questions in Table 6. IEEE Access, 9:70080–70094, 2021.
Practically speaking, this systematic review serves as a guide [6] Suhong Kim, Param Joshi, Parminder Singh Kalsi, and
for various researchers, practitioners, first responders, and Pooya Taheri. Crime analysis through machine learn-
crime analysts, to take into account the studied features and ing. In 2018 IEEE 9th Annual Information Technology,
techniques to effectively understand and detect the crimes Electronics and Mobile Communication Conference
that, in part, foster the effort for early crime detection. (IEMCON), pages 415–420. IEEE, 2018.
[7] Dewan Mamun Raza and Debasish Bhattacharjee Vic-
VII. CONCLUSION tor. Data mining and region prediction based on crime
The complexity of crimes has increased along with techno- using random forest. In 2021 International Conference
logical development, creating difficult problems for law en- on Artificial Intelligence and Smart Systems (ICAIS),
forcement. Researchers’ interest in utilizing machine learn- pages 980–987. IEEE, 2021.
ing and deep learning to predict crime has increased recently, [8] Lavanya Elluri, Varun Mandalapu, and Nirmalya Roy.
with an emphasis on finding patterns and trends in crime Developing machine learning based predictive models
occurrences. In order to analyze the various machine learning for smart policing. In 2019 IEEE International Con-
and deep learning algorithms used in predicting crime, this ference on Smart Computing (SMARTCOMP), pages
paper looks at more than 150 articles. We have significantly 198–204. IEEE, 2019.
studied the selected 51 articles to extract the essence of [9] Albert Meijer and Martijn Wessels. Predictive policing:
utilized various ML and DL techniques along with the pub- Review of benefits and drawbacks. International Jour-
licly available datasets. The use of machine learning and nal of Public Administration, 42(12):1031–1039, 2019.
deep learning algorithms to anticipate or identify criminal [10] Sohrab Hossain, Ahmed Abtahee, Imran Kashem, Mo-
activity has shown significant promise in resolving the crime hammed Moshiul Hoque, and Iqbal H Sarker. Crime
detection problem. These advances may help to increase the prediction using spatio-temporal data. In Computing
precision and efficacy of crime prediction models by lever- Science, Communication and Security: First Interna-
aging large datasets and sophisticated algorithms. Although tional Conference, COMS2 2020, Gujarat, India, March
there is a lack of literary wisdom on how these technologies 26–27, 2020, Revised Selected Papers 1, pages 277–
can be used to solve the problem of crime prediction, despite 289. Springer, 2020.
the advancements in this sector. Thus our findings help to un- [11] Miguel Saraiva, Irina Matijošaitienė, Saloni Mishra,
derstand the implications of various ML and DL techniques. and Ana Amante. Crime prediction and monitoring
Also, our mentioned datasets and future directions will help in porto, portugal, using machine learning, spatial and
the existing research community to pursue their research in text analytics. ISPRS International Journal of Geo-
the area of crime prediction. Information, 11(7):400, 2022.
[12] Ourania Kounadi, Alina Ristea, Adelson Araujo, and
REFERENCES Michael Leitner. A systematic review on spatial crime
[1] Neil Shah, Nandish Bhagat, and Manan Shah. Crime forecasting. Crime science, 9:1–22, 2020.
forecasting: a machine learning and computer vision [13] Locke J Morrisey. Bibliometric and bibliographic anal-
approach to crime prediction and prevention. Visual ysis in an era of electronic scholarly communication. In
Computing for Industry, Biomedicine, and Art, 4:1–14, Scholarly Communication in Science and Engineering
2021. Research in Higher Education, pages 149–160. Rout-
14 VOLUME 4, 2023

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

ledge, 2013. [26] Xinge Han, Xiaofeng Hu, Huanggang Wu, Bing Shen,
[14] Markus Hofmann and Andrew Chisholm. Text mining and Jiansong Wu. Risk prediction of theft crimes in
and visualization: Case studies using open-source tools, urban communities: an integrated model of lstm and st-
volume 40. CRC Press, 2016. gcn. IEEE Access, 8:217222–217230, 2020.
[15] P Tamilarasi and R Uma Rani. Diagnosis of crime rate [27] Zhonghang Li, Chao Huang, Lianghao Xia, Yong
against women using k-fold cross validation through Xu, and Jian Pei. Spatial-temporal hypergraph self-
machine learning. In 2020 fourth international confer- supervised learning for crime prediction. In 2022 IEEE
ence on computing methodologies and communication 38th International Conference on Data Engineering
(ICCMC), pages 1034–1038. IEEE, 2020. (ICDE), pages 2984–2996. IEEE, 2022.
[16] Akash Kumar, Aniket Verma, Gandhali Shinde, Yash [28] Nowshin Tasnim, Iftekher Toufique Imam, and MMA
Sukhdeve, and Nidhi Lal. Crime prediction using k- Hashem. A novel multi-module approach to predict
nearest neighboring algorithm. In 2020 International crime based on multivariate spatio-temporal data using
conference on emerging trends in information technol- attention and sequential fusion model. IEEE Access,
ogy and engineering (IC-ETITE), pages 1–4. IEEE, 10:48009–48030, 2022.
2020. [29] Binbin Zhou, Longbiao Chen, Sha Zhao, Shijian Li,
[17] Shubham Agarwal, Lavish Yadav, and Manish K Zengwei Zheng, and Gang Pan. Unsupervised domain
Thakur. Crime prediction based on statistical models. adaptation for crime risk prediction across cities. IEEE
In 2018 Eleventh International Conference on Contem- Transactions on Computational Social Systems, 2022.
porary Computing (IC3), pages 1–3. IEEE, 2018. [30] Umair Muneer Butt, Sukumar Letchmunan,
[18] Shraddha Ramdas Bandekar and C Vijayalakshmi. De- Fadratul Hafinaz Hassan, Mubashir Ali, Anees Baqir,
sign and analysis of machine learning algorithms for the Tieng Wei Koh, and Hafiz Husnain Raza Sherazi.
reduction of crime rates in india. Procedia Computer Spatio-temporal crime predictions by leveraging
Science, 172:122–127, 2020. artificial intelligence for citizens security in smart
[19] Ankit Sangani, Chirag Sampat, and Vijaya Pinjarkar. cities. IEEE Access, 9:47516–47529, 2021.
Crime prediction and analysis. In 2nd International [31] Shuyu Yao, Ming Wei, Lingyu Yan, Chunzhi Wang,
Conference on Advances in Science & Technology Xinhua Dong, Fangrui Liu, and Ying Xiong. Prediction
(ICAST), 2019. of crime hotspots based on spatial factors of random
[20] B Sivanagaleela and S Rajesh. Crime analysis and forest. In 2020 15th International Conference on Com-
prediction using fuzzy c-means algorithm. In 2019 3rd puter Science & Education (ICCSE), pages 811–815.
International Conference on Trends in Electronics and IEEE, 2020.
Informatics (ICOEI), pages 595–599. IEEE, 2019. [32] Mithileysh Sathiyanarayanan, Aisha Kanwal Junejo,
[21] A Mary Shermila, Amrith Basil Bellarmine, and Nir- and Odunayo Fadahunsi. Visual analysis of predictive
mala Santiago. Crime data analysis and prediction of policing to improve crime investigation. In 2019 Inter-
perpetrator identity using machine learning approach. national Conference on contemporary Computing and
In 2018 2nd international conference on trends in elec- Informatics (IC3I), pages 197–203. IEEE, 2019.
tronics and informatics (ICOEI), pages 107–114. IEEE, [33] Adelson Araújo, Nélio Cacho, Leonardo Bezerra, Car-
2018. los Vieira, and Julio Borges. Towards a crime
[22] Charlie Catlett, Eugenio Cesario, Domenico Talia, and hotspot detection framework for patrol planning. In
Andrea Vinci. A data-driven approach for spatio- 2018 IEEE 20th International Conference on High
temporal crime predictions in smart cities. In 2018 Performance Computing and Communications; IEEE
IEEE International Conference on Smart Computing 16th International Conference on Smart City; IEEE
(SMARTCOMP), pages 17–24. IEEE, 2018. 4th International Conference on Data Science and
[23] Charlie Catlett, Eugenio Cesario, Domenico Talia, and Systems (HPCC/SmartCity/DSS), pages 1256–1263.
Andrea Vinci. Spatio-temporal crime predictions in IEEE, 2018.
smart cities: A data-driven approach and experiments. [34] Abrar A Almuhanna, Marwa M Alrehili, Samah H
Pervasive and Mobile Computing, 53:62–74, 2019. Alsubhi, and Liyakathunisa Syed. Prediction of crime
[24] Fei Yi, Zhiwen Yu, Fuzhen Zhuang, Xiao Zhang, and in neighbourhoods of new york city using spatial data
Hui Xiong. An integrated model for crime prediction analysis. In 2021 1st International conference on ar-
using temporal and spatial factors. In 2018 IEEE In- tificial intelligence and data analytics (CAIDA), pages
ternational Conference on Data Mining (ICDM), pages 23–30. IEEE, 2021.
1386–1391. IEEE, 2018. [35] Anees Baqir, Sami ul Rehman, Sayyam Malik, Faizan
[25] Saroj Kumar Dash, Ilya Safro, and Ravi- ul Mustafa, and Usman Ahmad. Evaluating the per-
sutha Sakrepatna Srinivasamurthy. Spatio-temporal formance of hierarchical clustering algorithms to detect
prediction of crimes using network analytic approach. spatio-temporal crime hot-spots. In 2020 3rd Inter-
In 2018 IEEE International Conference on Big Data national Conference on Computing, Mathematics and
(Big Data), pages 1912–1917. IEEE, 2018. Engineering Technologies (iCoMET), pages 1–5. IEEE,
VOLUME 4, 2023 15

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

2020. work based on spatial analysis. Procedia computer

[36] Amal Algefes, Nouf Aldossari, Fatma Masmoudi, and science, 132:696–705, 2018.
Elham Kariri. A text-mining approach for crime tweets [47] A Wolf, TR Fanshawe, Amir Sariaslan, R Cornish,
in saudi arabia: from analysis to prediction. In 2022 7th Henrik Larsson, and S Fazel. Prediction of violent
International Conference on Data Science and Machine crime on discharge from secure psychiatric hospitals:
Learning Applications (CDMA), pages 109–114. IEEE, A clinical prediction rule (fovox). European psychiatry,
2022. 47:88–93, 2018.
[37] SPCW Sandagiri, BTGS Kumara, and Banujan [48] Kishan Bhushan Sahay, Bhuvaneswari Balachander,
Kuhaneswaran. Detecting crime related twitter posts B Jagadeesh, G Anand Kumar, Ravi Kumar, and
using artificial neural networks based approach. In 2020 L Rama Parvathy. A real time crime scene intelligent
20th International Conference on Advances in ICT for video surveillance systems in violence detection frame-
Emerging Regions (ICTer), pages 5–10. IEEE, 2020. work using deep learning techniques. Computers and
[38] M Anton Permana, Muhammad Ikhsan Thohir, Teddy Electrical Engineering, 103:108319, 2022.
Mantoro, and Media Anugerah Ayu. Crime rate detec- [49] Pradita Eko Prasetyo Utomo et al. Prediction the crime
tion based on text mining on social media using logistic motorcycles of theft using arimax-tfm with single input.
regression algorithm. In 2021 IEEE 7th International In 2018 Third International Conference on Informatics
Conference on Computing, Engineering and Design and Computing (ICIC), pages 1–7. IEEE, 2018.
(ICCED), pages 1–6. IEEE, 2021. [50] Varvara Ingilevich and Sergey Ivanov. Crime rate pre-
[39] Xiao Zhou, Xiao Wang, Gavin Brown, Chengchen diction in the urban environment using social factors.
Wang, and Peter Chin. Mixed spatio-temporal neu- Procedia Computer Science, 136:472–478, 2018.
ral networks on real-time prediction of crimes. In [51] Andrio Rodrigo Corrêa da Silva, Iális Cavalcante
2021 20th IEEE International Conference on Machine de Paula Júnior, Ticiana Linhares Coelho da Silva,
Learning and Applications (ICMLA), pages 1749– José Antônio Fernandes de Macêdo, and Wellington
1754. IEEE, 2021. Clay Porcino Silva. Prediction of crime location in
[40] Rajat Shenoy, Deepak Yadav, Harshita Lakhotiya, and a brazilian city using regression techniques. In 2020
Jignesh Sisodia. An intelligent framework for crime IEEE 32nd International Conference on Tools with
prediction using behavioural tracking and motion anal- Artificial Intelligence (ICTAI), pages 331–336. IEEE,
ysis. In 2022 International Conference on Emerging 2020.
Smart Computing and Informatics (ESCI), pages 1–6. [52] Priyanka Das, Asit Kumar Das, Janmenjoy Nayak,
IEEE, 2022. Danilo Pelusi, and Weiping Ding. Incremental clas-
[41] Nouf Aldossari, Amal Algefes, Fatma Masmoudi, and sifier in crime prediction using bi-objective particle
Elham Kariri. Data science approach for crime analysis swarm optimization. Information Sciences, 562:279–
and prediction: Saudi arabia use-case. In 2022 Fifth 303, 2021.
International Conference of Women in Data Science at [53] Zhongzhen Yan, Hao Chen, Xinhua Dong, Kewei Zhou,
Prince Sultan University (WiDS PSU), pages 20–25. and Zhigang Xu. Research on prediction of multi-
IEEE, 2022. class theft crimes by an optimized decomposition and
[42] Yujunrong Ma, Kiminori Nakamura, Eung-Joo Lee, fusion method based on xgboost. Expert Systems with
and Shuvra S Bhattacharyya. Eadtc: An approach to Applications, 207:117943, 2022.
interpretable and accurate crime prediction. In 2022 [54] Xu Zhang, Lin Liu, Minxuan Lan, Guangwen Song,
IEEE International Conference on Systems, Man, and Luzi Xiao, and Jianguo Chen. Interpretable machine
Cybernetics (SMC), pages 170–177. IEEE, 2022. learning models for crime prediction. Computers, En-
[43] Mohammed Boukabous and Mostafa Azizi. Multi- vironment and Urban Systems, 94:101789, 2022.
modal sentiment analysis using audio and text for crime [55] ML Trinhammer, AC Holst Merrild, JF Lotz, and
detection. In 2022 2nd International Conference on G Makransky. Predicting crime during or after psy-
Innovative Research in Applied Science, Engineering chiatric care: Evaluating machine learning for risk as-
and Technology (IRASET), pages 1–5. IEEE, 2022. sessment using the danish patient registries. Journal of
[44] Luiz GA Alves, Haroldo V Ribeiro, and Francisco A psychiatric research, 152:194–200, 2022.
Rodrigues. Crime prediction through urban metrics and [56] Asit Kumar Das and Priyanka Das. Graph based ensem-
statistical learning. Physica A: Statistical Mechanics ble classification for crime report prediction. Applied
and its Applications, 505:435–443, 2018. Soft Computing, 125:109215, 2022.
[45] Jingyi He and Hao Zheng. Prediction of crime [57] Weichao Liang, Youquan Wang, Haicheng Tao, and Jie
rate in urban neighborhoods based on machine learn- Cao. Towards hour-level crime prediction: A neural
ing. Engineering Applications of Artificial Intelligence, attentive framework with spatial–temporal-categorical
106:104460, 2021. fusion. Neurocomputing, 486:286–297, 2022.
[46] Hitesh Kumar Reddy ToppiReddy, Bhavna Saini, and [58] Umadevi V Navalgund and K Priyadharshini. Crime
Ginika Mahajan. Crime prediction & monitoring frame- intention detection system using deep learning. In 2018
16 VOLUME 4, 2023

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

International Conference on Circuits and Systems in uation. ACM SIGKDD Explorations Newsletter,
Digital Enterprise Technology (ICCSDET), pages 1–6. 22(1):18–33, 2020.
IEEE, 2018. [70] Ethan Carter, Tony Ward, and Annalisa Strauss-
[59] Mohammad Nakib, Rozin Tanvir Khan, Md Sakibul Hughes. The classification of crime and its related prob-
Hasan, and Jia Uddin. Crime scene prediction by lems: A pluralistic approach. Aggression and violent
detecting threatening objects using convolutional neural behavior, 59:101440, 2021.
network. In 2018 International Conference on Com- [71] Rashida Richardson, Jason M Schultz, and Kate Craw-
puter, Communication, Chemical, Material and Elec- ford. Dirty data, bad predictions: How civil rights vio-
tronic Engineering (IC4ME2), pages 1–4. IEEE, 2018. lations impact police data, predictive policing systems,
[60] Sharmila Chackravarthy, Steven Schmitt, and Li Yang. and justice. NYUL Rev. Online, 94:15, 2019.
Intelligent crime anomaly detection in smart cities us- [72] Peter M Asaro. Ai ethics in predictive policing: From
ing deep learning. In 2018 IEEE 4th International models of threat to an ethics of care. IEEE Technology
Conference on Collaboration and Internet Computing and Society Magazine, 38(2):40–53, 2019.
(CIC), pages 399–404. IEEE, 2018. [73] Oskar Josef Gstrein, Anno Bunnik, and Andrej Zwitter.
[61] Andrej Karpathy, George Toderici, Sanketh Shetty, Ethical, legal and social challenges of predictive polic-
Thomas Leung, Rahul Sukthankar, and Li Fei-Fei. ing. Católica Law Review, Direito Penal, 3(3):77–98,
Large-scale video classification with convolutional neu- 2019.
ral networks. In Proceedings of the IEEE conference on [74] Kiana Alikhademi, Emma Drobina, Diandra Prioleau,
Computer Vision and Pattern Recognition, pages 1725– Brianna Richardson, Duncan Purves, and Juan E
1732, 2014. Gilbert. A review of predictive policing from the
[62] Dingqi Yang, Terence Heaney, Alberto Tonon, Leye perspective of fairness. Artificial Intelligence and Law,
Wang, and Philippe Cudré-Mauroux. Crimetelescope: pages 1–17, 2022.
crime hotspot prediction based on urban and social [75] Michael Tonry. Predictions of dangerousness in sen-
media data fusion. World Wide Web, 21:1323–1347, tencing: Déjà vu all over again. Crime and Justice,
2018. 48(1):439–482, 2019.
[63] Alina Ristea, Mohammad Al Boni, Bernd Resch, [76] Rainer Mühlhoff. Predictive privacy: towards an ap-
Matthew S Gerber, and Michael Leitner. Spatial crime plied ethics of data analytics. Ethics and Information
distribution and prediction for sporting events using Technology, 23(4):675–690, 2021.
social media. International Journal of Geographical [77] Tzu-Wei Hung and Chun-Ping Yen. On the person-
Information Science, 34(9):1708–1739, 2020. based predictive policing of ai. Ethics and Information
[64] M Muthamizharasan and R Ponnusamy. Forecasting Technology, 23:165–176, 2021.
crime event rate with a cnn-lstm model. In Innova- [78] David Leslie. Understanding bias in facial recognition
tive Data Communication Technologies and Applica- technologies. Leslie, D.(2020). Understanding bias
tion: Proceedings of ICIDCA 2021, pages 461–470. in facial recognition technologies: an explainer. The
Springer, 2022. Alan Turing Institute. https://doi. org/10.5281/zenodo,
[65] Bao Wang, Penghang Yin, Andrea Louise Bertozzi, 4050457, 2020.
P Jeffrey Brantingham, Stanley Joel Osher, and Jack [79] Iqbal H Sarker. Machine learning: Algorithms, real-
Xin. Deep learning for real-time crime forecasting world applications and research directions. SN com-
and its ternarization. Chinese Annals of Mathematics, puter science, 2(3):160, 2021.
Series B, 40(6):949–966, 2019. [80] Sharad Goel, Ravi Shroff, Jennifer Skeem, and Christo-
[66] P William, Anurag Shrivastava, N Shun- pher Slobogin. The accuracy, equity, and jurisprudence
muga Karpagam, TA Mohanaprakash, Korakod of criminal risk assessment. In Research handbook on
Tongkachok, and Keshav Kumar. Crime analysis using big data law, pages 9–28. Edward Elgar Publishing,
computer vision approach with machine learning. In 2021.
Mobile Radio Communications and 5G Networks: [81] Romika Yadav and Savita Kumari Sheoran. Crime
Proceedings of Third MRCN 2022, pages 297–315. prediction using auto regression techniques for time
Springer, 2023. series data. In 2018 3rd International Conference and
[67] Caroline Wang, Bin Han, Bhrij Patel, and Cynthia Workshops on Recent Advances and Innovations in
Rudin. In pursuit of interpretable, fair and accurate Engineering (ICRAIE), pages 1–5. IEEE, 2018.
machine learning for criminal recidivism prediction. [82] Hongning Wang and Sanjun Ma. Preventing crimes
Journal of Quantitative Criminology, pages 1–63, 2022. against public health with artificial intelligence and ma-
[68] Julia Dressel and Hany Farid. The dangers of risk chine learning capabilities. Socio-Economic Planning
prediction in the criminal justice system. 2021. Sciences, 80:101043, 2022.
[69] Raha Moraffah, Mansooreh Karami, Ruocheng Guo, [83] Joshua Abraham, Ronnie Ng, Marie Morelato, Mark
Adrienne Raglin, and Huan Liu. Causal interpretabil- Tahtouh, and Claude Roux. Automatically classifying
ity for machine learning-problems, methods and eval- crime scene images using machine learning methodolo-
VOLUME 4, 2023 17

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

gies. Forensic Science International: Digital Investiga-

tion, 39:301273, 2021.

18 VOLUME 4, 2023

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

APPENDIX. ADDITIONAL CRIME DETECTION PAPERS

USING MACHINE LEARNING AND DEEP LEARNING

TABLE 7: Crime detection papers with data science algorithms

Method Algorithms Applied Data Set Performance

Applied six different types of machine learning 2001 to 2012 with
NB, CART, KNN, LoR Aggregate Accuracy
algorithms using similar characteristics on crime more than 2,00,000
and SVM - 30%
data. [15] records
Police website of a
K-nearest neighbor algorithm is applied to Aggregate Accuracy
KNN city Indore in
predcit the crime. [16] - 99.51%
Madhya Pradesh
Weighted Moving
Statistical modeling approaches are applied to Average, Functional
Crimes in India 2001 Between the range
predict the crime rate of a particular district of Coefficient Regression
- 2013 85% and 90%
India using the previous year’s statistics. [17] and Arithmetic
Geometric Progression
Predicts the description of the perpetrator using San Francisco
algorithms like Multilinear Regression, K nearest KNN Homicide dataset 91.6%
neighbors, and Neural Network. [21] (1981-2014)
Machine-Learning predictive models, K-nearest Vancouver crime
KNN and boosted Accuracy between
neighbor, and boosted decision trees are data for the last 15
decision tree 39% to 44%
implemented to predict crime. [6] years
Spatial analysis and auto-regressive models to Association rule mining
Mean Absolute Error
detect high-risk crime regions in urban areas and , classification, and Chicago
(MAE) - 11.5
reliably forecast crime trends in each region. [22] clustering
Gaussian Process
Regressor, RF, Auto
Regression Moving
An auto-regressed temporal and a feature-based
Average, Linear Chicago from Jan. 1 Aggregate RMSE -
inter-area spatial correlation are applied to
Regression and 2013 to Jan. 1 2016. 3.1
measure such patterns for crime prediction. [24]
Clustered Continuous
Conditional Random
Field
Experimented with polynomial, auto-regressive, Polynomial,
and support vector regression methods and found auto-regressive and Different regions in RMSE- 0-10 for
that the quality of support vector regression support vector City of Chicago most crimes
significantly outperforms other approaches.[25] regression
Representative covariates from the nonhistorical
crime data are added to the prediction model to Crime data set in San Demo, POI, DI -
Random Forest
explore the changes in the result accuracy of Francisco 1.18
crime prediction. [31]
Hierarchical
Density-based spatial
Spatio-Temporal Crime Hot-Spots detection Approx 10 sec for
clustering of NYC
based on clustering method. [35] cluster
applications with noise
(HDBSCAN)
Check crime rate based on text mining on social Aggregate Accuracy
Logistic Regression Twitter & Facebook
media using a logistic regression algorithm. [38] - 90%
Latitude and Longitude are used by KNN. In
KNN & Naive Bayes UK crime data (2015 Accuracy - 70 to
addition, Date is used by the naive bayes
(NB) to 2017). 80%
algorithm for classifying crime type. [46]
A Text-mining approach for crime tweets in CNN, KNN, NB, SGD, Aggregate Accuracy
Tweets
Saudi Arabia. [36] SVM, DT and RF - 79%

VOLUME 4, 2023 19

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

Presents an empirical analysis of machine

learning algorithms for crime prediction using 2001-2015 Crime
DT, RF, SVM and KNN Accuracy - 99.5%
stacked generalization, which is an ensemble Dataset in India
approach. [20]
Proposes using auto-regression techniques for ARIMA and
Crime Data set from
predicting crime in a city using time series data. Generalized Linear Not Quantified
India
[81] Model
Proposes a Spatial-Temporal Self-Supervised
ST-HSLARIMA, SVM,
Hypergraph Learning framework (ST-HSL) to NYC and Chicago MAE 0.79 for
STResNet, DCRNN,
tackle the label scarcity issue in crime prediction. Crime data robbery
STGCN and GWN
[27]
Proposes an artificial neural network (ANN)
Neural Network and
based approach for detecting crime-related posts Tweets Accuracy - 90.33%
SVM
on Twitter. [37]
Records of crimes
Machine learning algorithms to predict crimes against public health
SVM and Random
that endanger public health and aim to improve in a chinese city Not Quantified
Forest
the efficiency of crime prevention. [82] from January 1, 2018
to October 31, 2018.
2248 individuals
Develop scalable predictive models for violent
discharged from Aggregate
offending following (12 to 24 months
Cox Regression Swedish psychiatric Concordance Index
post-discharge) discharge from secure psychiatric
hospitals between 0.73
hospitals. [47]
1992 and 2013
Application of two different machine learning Australian Federal
classifier models on a rich collection of SVM with Bag of Visual Police illicit drug True positive rate -
real-world forensic casework images of words and Tree-CNN database. 97287 89.17%
drug-related offenses. [83] images
Investigates the HDBSCAN to detect the hot
spots that have a higher risk of crime occurrence
and Seasonal Auto-Regressive Integrated Moving
HDBSCAN and NYC crimes 2008 to Aggregate MAE -
Average (SARIMA) is exploited in each dense
SARIMA 2017. 11.7
crime region to predict the number of crime
incidents in the future with spatial and temporal
information. [30]

20 VOLUME 4, 2023

Varun et al.: Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

VARUN MANDALAPU graduated with a NIRMALYA ROY is currently a Professor in the

Ph.D from the University of Maryland Baltimore Department of Information Systems at University
County in the Department of Information Systems of Maryland Baltimore County. He directs the
with a specialization in Artificial Intelligence & Mobile, Pervasive and Sensor Computing (MPSC)
Knowledge Management. Varun currently works Lab. He was a Clinical Assistant Professor in the
as a Senior Data Scientist at Mutual of Omaha School of Electrical Engineering and Computer
and earlier worked as a research assistant under Science at Washington State University from Jan-
Dr. Jiaqi Gong at Sensor Accelerated Intelligent uary 2012 to June 2013. Prior to that, he worked
Learning (SAIL) Lab and co-advised by Dr. as a Research Scientist at Institute for Infocomm
Zhiyuan Chen and Dr. Karen Chen. Before en- Research (I2R), Singapore from 2010 to 2011. He
rolling in Ph.D. degree, Varun completed his master’s degree in Management was a postdoctoral fellow in Electrical and Computer Engineering Depart-
Information Systems at the University of Illinois Springfield and a master’s ment at The University of Texas at Austin from 2008 to 2009. He received his
degree in Sensor Systems Technology from Vellore Institute of Technology, Ph.D. and M.S. in Computer Science and Engineering from The University
Vellore, India. Varun is also an affiliate member of IEEE EMBS technical of Texas at Arlington in 2008 and 2004 respectively. He did his Bachelors
committee on Wearable Biomedical Sensors and Systems. His research in Computer Science and Engineering from Jadavpur University, India in
publications appeared in reputed AI venues such as AAAI Workshops, 2001.
Artificial Intelligence in Education, Educational Data Mining, Smart Health,
IEEE Body Sensor Networks, and IEEE Biomedical Health Informatics.

LAVANYA ELLURI ( Member, IEEE) She is an

Assistant Professor of Computer Information Sys-
tems at the Texas A&M University-Central Texas.
She received Ph.D. in Information Systems from
the University of Maryland Baltimore County and
a Master of Science in Management Information
Systems from the University of Houston Clear
Lake. She has worked for over a decade in the
IT industry at reputed companies Infosys and REI
Systems. She has led several projects at REI Sys-
tems and has extensive work experience with various databases and data
warehousing technologies. Also, she has experience working with a wide
range of data science and data analytics projects. Her research and teaching
interests include data analytics, data science, semantic web, database sys-
tems, data privacy and security, text mining, and healthcare IT systems. Her
research publications appeared in reputed venues such as IEEE Big data,
IEEE cloud, IEEE Access, and Frontiers in Bigdata.

PIYUSH VYAS is an Assistant Professor of Com-

puter Information Systems at the Texas A&M
University-Central Texas.He received the Ph.D.
and MS degree in Information Systems from
Dakota State University, Madison, SD, USA in
2022 and 2020 respectively. He received a Master
of Engineering and Bachelor of Engineering de-
gree in Information Technology from State Tech-
nical University Bhopal India, in 2012 and 2009
respectively. His current research interests include
text mining, association rule mining, Traditional and Online machine learn-
ing, transfer learning, deep learning, and data mining in the domain of eCom-
merce, social media, microblogging, healthcare, medical imaging, the dark
web, and explainable AI. He has published his articles in IEEE Transactions
of Technology and society, Issues of information systems, International Jour-
nal of Information Security and Privacy (IJISP)-IGI Global, AIS-AMCIS,
AIS-MWAIS, DSI, and IEEE conferences. He has received the best research
paper award for AMCIS 2021. His teaching interest is machine learning, data
communications, computer networks, business analytics, system analysis
and design, and database management systems.

VOLUME 4, 2023 21

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Fbi Crime Analysis and Prediction Using Machine Learning
No ratings yet
Fbi Crime Analysis and Prediction Using Machine Learning
8 pages
24 M Crime Prediction Using Machine Learning and Deep Learning a Systematic Review and Future Directions
No ratings yet
24 M Crime Prediction Using Machine Learning and Deep Learning a Systematic Review and Future Directions
18 pages
Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions
No ratings yet
Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions
35 pages
Forecasting Criminal Activity Using Machine Learning Approaches
No ratings yet
Forecasting Criminal Activity Using Machine Learning Approaches
7 pages
IRJET-V11I4287
No ratings yet
IRJET-V11I4287
6 pages
Crime Examination Study 2021
No ratings yet
Crime Examination Study 2021
9 pages
Evolving Data Mining Algorithms On The Prevailing Crime Trend - An Intelligent Crime Prediction Model
No ratings yet
Evolving Data Mining Algorithms On The Prevailing Crime Trend - An Intelligent Crime Prediction Model
6 pages
Crime Prediction Using Machine Learning and Deep L
No ratings yet
Crime Prediction Using Machine Learning and Deep L
8 pages
Research Paper
No ratings yet
Research Paper
11 pages
272crime Rate Prediction Using Machine Learning
No ratings yet
272crime Rate Prediction Using Machine Learning
5 pages
abcde
No ratings yet
abcde
5 pages
Arabia Saudita
No ratings yet
Arabia Saudita
6 pages
computers-12-00255
No ratings yet
computers-12-00255
28 pages
Crime Prediction and Analysis Using Data Mining
No ratings yet
Crime Prediction and Analysis Using Data Mining
6 pages
Crime Prediction and Analysis: 1 Pratibha 2 Akanksha Gahalot
No ratings yet
Crime Prediction and Analysis: 1 Pratibha 2 Akanksha Gahalot
6 pages
Paper (Imran)
No ratings yet
Paper (Imran)
13 pages
Predicting Violent Crime Hot-Spots Utilizing Machine Learning
No ratings yet
Predicting Violent Crime Hot-Spots Utilizing Machine Learning
9 pages
The Use of Data Mining Techniques in Crime Prevention and Prediction-Not Good
No ratings yet
The Use of Data Mining Techniques in Crime Prevention and Prediction-Not Good
4 pages
Crime Hot Spot Prediction-1
No ratings yet
Crime Hot Spot Prediction-1
7 pages
IRJET-V10I457
No ratings yet
IRJET-V10I457
4 pages
Crime Analysis Surveypaper
100% (1)
Crime Analysis Surveypaper
6 pages
Crime Analysis and Prediction Using Data
No ratings yet
Crime Analysis and Prediction Using Data
7 pages
journal_paper
No ratings yet
journal_paper
3 pages
Exploratory Data Analysis and Crime Prevention Using Machine Learning The Case of Ghana
No ratings yet
Exploratory Data Analysis and Crime Prevention Using Machine Learning The Case of Ghana
10 pages
Dataminig in crimerate
No ratings yet
Dataminig in crimerate
9 pages
Crime Forecasting
No ratings yet
Crime Forecasting
14 pages
AbhayRautela_MiniProject_5th Semester
No ratings yet
AbhayRautela_MiniProject_5th Semester
15 pages
Fyp Caps
No ratings yet
Fyp Caps
24 pages
22df PDF
No ratings yet
22df PDF
4 pages
crime in ml
No ratings yet
crime in ml
10 pages
Crime Rate Prediction: Ch. Mahendra1, G. Nani Babu2, G. Balu Nitin Chandra, A. Avinash 4, Y. Aditya5
No ratings yet
Crime Rate Prediction: Ch. Mahendra1, G. Nani Babu2, G. Balu Nitin Chandra, A. Avinash 4, Y. Aditya5
6 pages
Research_paper_conference_(1)_(1)[1]
No ratings yet
Research_paper_conference_(1)_(1)[1]
4 pages
Enhanced_Crime_Prediction_Project_Report
No ratings yet
Enhanced_Crime_Prediction_Project_Report
4 pages
Suraksha Kavach Crime Prediction and Analysis
No ratings yet
Suraksha Kavach Crime Prediction and Analysis
13 pages
SurveyPaperFinal
No ratings yet
SurveyPaperFinal
4 pages
Crime - Data-Mining-And-K-Means 2018
No ratings yet
Crime - Data-Mining-And-K-Means 2018
4 pages
Criminalistics Proposal
No ratings yet
Criminalistics Proposal
13 pages
Crime Rate Prediction Using KNN: Ms. Vrushali Pednekar Ms. Trupti Mahale Ms. Pratiksha Gadhave Prof. Arti Gore
No ratings yet
Crime Rate Prediction Using KNN: Ms. Vrushali Pednekar Ms. Trupti Mahale Ms. Pratiksha Gadhave Prof. Arti Gore
4 pages
JCP 01 00029
No ratings yet
JCP 01 00029
17 pages
Final RP PDF
No ratings yet
Final RP PDF
3 pages
IJSC Vol 7 Iss 4 Paper 5 1459 1466
No ratings yet
IJSC Vol 7 Iss 4 Paper 5 1459 1466
9 pages
Devil Crime Rate Prediction Using K-Means
No ratings yet
Devil Crime Rate Prediction Using K-Means
14 pages
(IJIT-V7I2P3) :CH - Kalpana, A. Sobhana Rhosaline
No ratings yet
(IJIT-V7I2P3) :CH - Kalpana, A. Sobhana Rhosaline
5 pages
Krithika Heheee
No ratings yet
Krithika Heheee
17 pages
Ijcsit 2021120201
No ratings yet
Ijcsit 2021120201
9 pages
Crime Analysis and Prediction Using Machine Learning
No ratings yet
Crime Analysis and Prediction Using Machine Learning
5 pages
AI Based Crime Rate Prediction
No ratings yet
AI Based Crime Rate Prediction
6 pages
Document 1
No ratings yet
Document 1
24 pages
Research-paper(changes)
No ratings yet
Research-paper(changes)
6 pages
An Analysis of Data Mining Applications in Crime Domain: P. Thongtae and S. Srisuk
No ratings yet
An Analysis of Data Mining Applications in Crime Domain: P. Thongtae and S. Srisuk
5 pages
Crime Prediction Using Machine Learning
No ratings yet
Crime Prediction Using Machine Learning
19 pages
A Survey of Data Mining Techniques For C
No ratings yet
A Survey of Data Mining Techniques For C
6 pages
Article3
No ratings yet
Article3
8 pages
Epistemologies of Predictive Policing
No ratings yet
Epistemologies of Predictive Policing
13 pages
Enhancements in The World of Digital Forensics
No ratings yet
Enhancements in The World of Digital Forensics
7 pages
TestEngineering
No ratings yet
TestEngineering
8 pages
Crime Predictionand Analysis
No ratings yet
Crime Predictionand Analysis
7 pages
Life and Science Journal
No ratings yet
Life and Science Journal
4 pages
1 s2.0 S0020025521001225 Main
No ratings yet
1 s2.0 S0020025521001225 Main
25 pages
Activity Recognition: Fundamentals and Applications
From Everand
Activity Recognition: Fundamentals and Applications
Fouad Sabry
No ratings yet
UCP Research Paper
No ratings yet
UCP Research Paper
14 pages
3 AI Policing in India Existing Research and Where To Begin Future Research
No ratings yet
3 AI Policing in India Existing Research and Where To Begin Future Research
10 pages
Research Paper B
No ratings yet
Research Paper B
44 pages
Dirty Data and Bad Prediction
No ratings yet
Dirty Data and Bad Prediction
30 pages
Regulating Artificial Intelligence For Use in Criminal Justice Systems Fair Trials
No ratings yet
Regulating Artificial Intelligence For Use in Criminal Justice Systems Fair Trials
37 pages
Algorithmic Decision Making - Case Study
No ratings yet
Algorithmic Decision Making - Case Study
34 pages
Preview-9780429555916 A41017822
No ratings yet
Preview-9780429555916 A41017822
34 pages
12104-Artificial-Intelligence-in-the-Criminal-Justice
No ratings yet
12104-Artificial-Intelligence-in-the-Criminal-Justice
7 pages
Social Impacts Case Studies
No ratings yet
Social Impacts Case Studies
6 pages
Enhancing Law Enforcement Through Information and Communication Technology
No ratings yet
Enhancing Law Enforcement Through Information and Communication Technology
3 pages
Predictive Policing
No ratings yet
Predictive Policing
12 pages
Lea 2 Midterm p1
No ratings yet
Lea 2 Midterm p1
6 pages
Get Encyclopedia of Community Policing and Problem Solving 1st Edition Kenneth J. Peak (Ed.) PDF Ebook With Full Chapters Now
100% (12)
Get Encyclopedia of Community Policing and Problem Solving 1st Edition Kenneth J. Peak (Ed.) PDF Ebook With Full Chapters Now
70 pages
Algorithms and Crime
No ratings yet
Algorithms and Crime
20 pages
AI Ethics Case Study - Edited
No ratings yet
AI Ethics Case Study - Edited
2 pages
Chat GPT Prompts
No ratings yet
Chat GPT Prompts
19 pages
Pro Active and Predictive Public Safety and Security
No ratings yet
Pro Active and Predictive Public Safety and Security
82 pages
Bias in Algorithms
No ratings yet
Bias in Algorithms
106 pages
Prelim Exam - Comparative Models in Policing
No ratings yet
Prelim Exam - Comparative Models in Policing
2 pages
Loopholes in The Fourth Amendment: Predictive Policing and The Threat of Data Brokers - Flora Tian
No ratings yet
Loopholes in The Fourth Amendment: Predictive Policing and The Threat of Data Brokers - Flora Tian
18 pages
Research 2024
No ratings yet
Research 2024
34 pages
Beyond Street Smarts: Data-Driven Crime Fighting
No ratings yet
Beyond Street Smarts: Data-Driven Crime Fighting
6 pages
Chapter 11
No ratings yet
Chapter 11
20 pages
Artificial Intelligence
No ratings yet
Artificial Intelligence
9 pages
Accounting, Finance, Sustainability, Governance & Fraud Theory and
100% (1)
Accounting, Finance, Sustainability, Governance & Fraud Theory and
231 pages
Ethical Considerations in Predictive Policing in Law Enforcement Practices
0% (1)
Ethical Considerations in Predictive Policing in Law Enforcement Practices
14 pages
Datajusticelab Critical Data Literacy Guidebook
No ratings yet
Datajusticelab Critical Data Literacy Guidebook
17 pages
Assessment 1 - Homicide Data Mining Presentation
No ratings yet
Assessment 1 - Homicide Data Mining Presentation
17 pages
SFPD PredPol Emails 2 PDF
No ratings yet
SFPD PredPol Emails 2 PDF
51 pages