Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1128

Machine Learning Approaches to Classification of

Online Users by Exploiting Information
Seeking Behaviours
Shaik. Allabhakshu1; Manam Om Rupesh 2; Kodela Jayasri3;
Thungaturthi Satya Sai Himaja4; Katikam Mahesh5
1UGStudent
Tirumala Institute of Technology & Sciences Satuluru, Nadendla (MD), Palnadu (DT).
2UGStudent
Tirumala Institute of Technology & Sciences Satuluru, Nadendla (MD), Palnadu (DT).
3UGStudent
Tirumala Institute of Technology & Sciences Satuluru, Nadendla (MD), Palnadu (DT)
4UGStudent
Tirumala Institute of Technology & Sciences Satuluru, Nadendla (MD), Palnadu (DT).
5
Assistant Professor, Tirumala Institute of Technology & Sciences Satuluru, Nadendla (MD), Palnadu (DT).

Publication Date: 2025/05/02

Abstract: In today’s digital age, understanding how users interact with online platforms has become more important than
ever, especially for reshape experiences and protect security. This project introduces an innovative approach to analyzing
and classifying user behavior by using machine learning, with a focus on predicting information-seeking patterns based on
social media and locating data. Inspired by real world needs, we developed a system that uses a fine-tuned Random Forest
Classifier to categorize user activities into "uncertain Behavior, Good Behavior, or Neutral Behavior using features like
gender, age, location latitude and longitude, and social metrics such as followers, friends, favorites, and statuses. The
model does a great job reaching an impressive accuracy of 90.21%. What makes this project special is its interactive edge
we built a user friendly interface using Jupyter allowing anyone to input their own data think of it like filling out a digital
profile and get instant predictions about their behavior type. It is for marketer wanting to personalize ads, security teams
detecting possible risk, or researcher studying online habits, this tool delivers action able insights with a simple click. The
system also save predictions to a CSV file for future reference and offers a peek into advanced possibilitie with plans for
real time deployment using Flask and Drawing from established research on user direction and machine learning, this
project balances technical culture with practical usability aiming to enhance our understanding of digital behavior while
keeping privacy and ethics in mind. It a step toward smarter more natural online environments crafted with care and
Interest.

Keywords: Machine learning, User behaviors, Random Forest Classifier, Accuracy, Interactive interface, Privacy, Social media,
Information-seeking patterns, Real-time deployment, User-friendly interface , Security K. Jayasri, SK. AllaBhakshu, M. Om
Rupesh, T. SatyaSaiHimaja, K. Mahesh, 2025, Machine Learning Approaches to classification of online users by exploiting
information seeking behaviour.

How to Cite: Shaik.Allabhakshu; ManamOmRupesh; KodelaJayasri; Thungaturthi Satya SaiHimaja; KatikamMahesh (2025).
Machine Learning Approaches to Classification of Online Users by Exploiting Information Seeking Behaviours.
International Journal of Innovative Science and Research Technology, 10(4), 2247-2252.
https://doi.org/10.38124/ijisrt/25apr1128

I. INTRODUCTION Center to this project is a Random Forest Classifier

that analyze features like gender, age, location latitude
In today internet influence world understanding online longitude, and social media metrix followers, friends,
user behavior is both and exciting. Every click and scroll favorite, statuses to classify users into Suspicious, Good, or
reveals whether users seek direction, transactions, or Neutral behavior types. Data preparation involves encoding
information. This project, Machine Learning Approaches to categorical variables and splitting it into training and testing
Classification of Online Users by Exploiting Information sets for effective learning. Achieving the model delivers
Seeking Behaviours uses advanced machine learning to reliable insights, supported by metrics like classification
decode these predicting behaviors like interest purchase reports and confusion matrices, vividly displayed through
intent Inspired by real world applications personalized heat maps and line graphs tracking follower status trends.
marketing, and interaction it aims to enhance user This project makes truly attractive is conjoinent component.
understanding with a human touch. We designed a user friendly interface using Jupyter tool,

IJISRT25APR1128 www.ijisrt.com 2247

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1128
grant anyone be a buyer, a security analystic, or a curious security, and research. It features a structured preprocessing
explorer to input their own data and receive instant pipeline with Label Encoder for categorical variables and an
predictions about their behavior type. Picture typing in your 80/20train test split, attaining 90.21%accuracy, validated by
details, hitting a "Predict" button, and watching the system metrics like accuracy scores and visualized through
reveal your digital persona with a vibrant display Looking Matplotlib line graphs and Seaborn Event tracking. The
forward , we excited about the potential to extend this into interactive Jupyter widget interface allows real-time input of
real-time applications using tools like Flask and Pyngrok, data and displays predictions in a styled output box, with
making it a living, breathing tool for the digital age. This results saved to "Predicted_Output.csv," while Flask and
project make on a of research into information-seeking Pyngrok support future real time deployment. Combining
behavior, from Information Foraging Theory to modern advanced analytics with access, this system offers a practical
deep learning advancements, while keeping an eye on tool for diverse users and sets the stage for enhancements
ethical considerations like privacy and consent. It’s more like multi model data.
than just a technical exercise it a bridge between data
science and human understanding, aiming to create smarter, IV. SYSTEM ARCHITECTURE
more intuitive online environments.
The system architecture for the Machine Learning
II. LITERATURE REVIEW Approaches to Classification of Online Users by Exploiting
Information Seeking Behaviours project is a modular
User behavior analysis and classify have totally, building framework that integrates data processing, machine
on foundational theories and machine learning techniques to learning, visualization, and inter activity to classify online
decode online direction. Pirolli & Card(1999) [1]introduced user behavior.
Information Foraging Theory, while Chooetal. (2000) [2]
grouping users into undirected, conditioned, informal, and Data Preprocessing Layer: This layer cleans and
formal search types, and White & Drucker (2007) [3] used prepares raw input data, including features like gender,
machine learning to identify navigational, transactional, and latitude, and social media metrics, by encoding categorical
informational query patterns. Liu et al. (2016) [4] applied variables with Label Encoder.
SVMs, Moreno & Redondo (2016) [5] used decision trees
and random forests for e commerce, and Kumar et al. (2018) Machine Learning Core: main of the system a well
[6] enhanced long term prediction with LSTMs. tuned Random Forest Classify with 200valuepredicts
Unsupervised learning including Zhang & Nasraoui’s Suspicious, Good, Neutral behaviors with 90.21% accuracy,
(2008) [7] K-Means for click stream connection and Chen et evaluated using accuracy score, grouping report, and
al.’s (2013) [8] DBSCAN for bot detection, along side confusion matrix to assess.
Chatzopoulou et al.’s (2010) [9] YouTube analysis, has been
pivotal. Deep learning advancements, Mnihetal.’s(2015) Visualization Layer: This layer enhances
[10] DQN for search reduce, Rendle et al.’s (2009) [11] understanding by employing Matplotlib to generate line
factorisation machines, and Huang et al.’s (2020) [12] graphs and Seaborn to create heatmaps, providing
difference coders for variation detection , have transformed perception into user trends and model efficacy.
the field. However, privacy concerns, highlighted by
Barocas & Nissenbaum(2014)[13]andShenet al.(2019)[14] Interactive Interface Layer: Built with Jupyter widgets,
watchful tracking. this user facing component allows input of data e.g Fid,
User Id via text fields and drop downs, processing it in real
III. PROPOSED METHODOLOGY time with a "Predict" button to display results in a styled
HTML.
The proposed system for the Machine Learning Approaches
to Classification of Online Users by Exploiting Information Data Storage Module : unmixed with the interface this
Seeking Behaviours project is a user central solution that module saves prediction outputs to Predicted_Output.csv,
uses a well tuned Random Forest Classifier to identifying enabling easy storage and analysis of results.
and classify online behavior into Suspicious, Good, or
"Neutral" types based on features like gender, age, location Deployment Laye: Designed for scalability, this layer
(latitude/longitude), and social media metrics (followers, uses Flask and Pyngrok for potential real time web
friends, favorites, statuses), addressing needs in marketing, deployment, to reduce performance.

IJISRT25APR1128 www.ijisrt.com 2248

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1128

Fig1 System Architecture.

IJISRT25APR1128 www.ijisrt.com 2249

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1128
This architecture ensures a smooth flow from data for particular applications.
ingestion to actionable insights, offering usability and
extensibility for future enhancement slik emultimodal data, User Friendly Interactivity: The interface, built with
while prioritizing privacy and ethical considerations in the Jupyter widgets, allows user to input data easily via
digital landscape. textfield, dropdown, providing immedent predictions in
output box, enhancing access for buyer, security team, and
V. PROPOSED ALGORITHMS EXPLANATION researcher.

 Random Forest Classifier: Comprehensive Data Insights: uptake tools like

Description: Uses an ensemble of decision trees from Matplotlib line graphs e.g follower vs. statuse and Seaborn
sklearn. ensemble to predict Suspicious, Good, or Neutral heat maps for confusion matrices offer into user trends and
behavior. model performance, decision making and pattern
recognition.
 Process:
Trains on preprocessed data, predicts with 90.21% Scalability and Future Ready: Integration of Flask and
accuracy, and handles features like gender and followers. Pyngrok lays the ground work for real time deployment,
with potential for cloud support and, ensuring the system
 Advantages: can scale to handle larger datasets or real world.
Reduces overfitting and scales well for buyers and
security applications. Practical Output Storage: The ability to save
predictions to "Predicted_Output.csv" provides a convenient
 Label Encoding : way to store and analyze results over time, supporting long
Description: Converts grouping variables to numbers term research and operational needs.
using sklearn. preprocessing.
Ethical and Privacy Consideration : research and
 Process: usability with privacy in mind, the system balances with
Encodes data before training and decodes predictions ethical standard, fostering trust among users.
for display.
VII. RESULTS
 Advantages:
Ensure speace with the classify. Train,Testsplit: The Machine Learning Approaches to Classification of
Online Users by Exploiting Information Seeking Behaviours
 Description: project impressive result , achieving a 90.21% accuracy
Splits data into training and testing sets from with the Random Forest Classify to grouping user into
sklearn.model selection. Suspicious, Good, or Neutral behaviors based on features
like gender, age, location, and social metrics, with specific
 Process : predictions such as Label:0|SuspiciousBehavior
Uses random state to assess model performance. forFid:172.217.3.106- 10.42.211-443.5123-6 and Label: 1 |
Advantages: Prevents overfitting and validates 90.21% Good Behavior, forFid:172.217.10.238-10.42.151.44
accuracy. 3.Itprovide

VI. ADVANTAGES OF THE PROPOSED detailed performance through classification report and
SYSTEM a Seaborn visualized confusion matrix, complemented by
Matplotlib line graphs showing trends like a peak at 500
High Accuracy and Reliability: The system achieves an followers with 20,000 statuses, while saving predictions to
90.21% accuracy with a fine tuned Random Forest "Predicted_Output.csv" for analysis and offering real-time
Classifier, classification of grouping of user behaviors usability via a Jupyter widget interface with styled HTML
Suspicious, Good, Neutral based on diverse features like outputs, demonstrating its reliability and practicality. As
gender, age, location, and standard, making it trust worthy showing figures below:

Fig 2 Entering the details.

IJISRT25APR1128 www.ijisrt.com 2250

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1128

Fig 3 Suspicious Behavior

Fig 4 Good Behavior.

VIII. CONCLUSION As of 09:47 PM PDT on Thursday, April 10, 2025, this

system not only deepen our understanding of digital
The Machine Learning Approaches to Classification of behavior but also seta foundation for future enhancements,
Online Users by Exploiting Information Seeking Behaviours solidifying its role as a valuable tool in the evolving online
project marks a successful end in leveraging advanced landscape.
machine learning to classify online user behavior with a
90.21% accuracy using a well tuned Random Forest To further demonstrate the system functionality, a
Classifiy By analysis features such as gender, age, location, sample of input data and the corresponding predicted
and social media metrics, it effectively grouping user into behavior type is presented below. The table highlight
Suspicious, Good or Neutral behavior , delivering actionable various user based on attributes such as gender, age,
insights for marketers, security teams, and researcher. The geographic location, and social media metrics like follower,
interactive Jupyter widget interface enhance usability with friend, favorite, and status. These inputs were processed
real-time prediction and shown output like heat map and line through the Random Forest Classifier, which then grouped
graph, while the ability to save result to the behavior as Suspicious, Good, or Neutral with an
"Predicted_Output.csv" add practical value. Drawing from accuracy of 90.21%.Table 1 illustrates how the model
established research and balancing technical with privacy interprets and classifies real world user data inputs into
consideration , the project demonstrate performance and behavioral groupings that lead to actions. As per the below
scalability potential through Flask and Pyngrok integration. table.

Table1Sample User Data and Predicted Behavior Type

User ID Gender Age Latitude Longitude Followers Friends Favorites Statuses Predicted Behavior
780123 Male 26 42.9572 -85.6869 89 241 251 837 Good Behavior
5123-6 Female 23 43.2287 -85.5602 89 251 251 837 Suspicious Behavior
979863 Male 29 40.7128 -74.0060 120 420 134 530 Suspicious Behavior
689754 Female 21 34.0522 -118.2437 500 620 345 20000 Good Behavior

IJISRT25APR1128 www.ijisrt.com 2251

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1128
ACKNOWLEDGMENT [10]. M.Soleymani,M.Riegler,andP.Halvorsen,‘‘Multimod
al analysis of user behavior and browsed
The authors are especially thankful to Mr. K. Mahesh, content under different imagesearch intents,’’
Assistant Professor, for his valuable guidance, consistent Int. J. Multimedia Inf. Retr., vol. 7, no. 1, pp.29–
motivation, and support at every stage of this work. His 41,Mar.2018,doi:10.1007/s13735-018-0150-6.
insights and expertise greatly contributed to the successful [11]. D. Koehn, S. Lessmann, and M. Schaal, ‘‘Predicting
completion of this project. online shopping behaviour from Click stream data
using deeplearning, ’’ExpertSyst. Appl., vol. 150,
We also acknowledge the use of open source tools and Jul. 2020, Art. no. 113342.
libraries such as Scikit-learn, Matplotlib, Seaborn, and [12]. H.Yoganarasimhan,‘‘Search personalization using
Jupyter, which made this research possible. This project has machine learning,’’ Manage. Sci., vol. 66,
been a right set of circumstance to enhance our no.3,pp.1045–1070,Mar.2020.
practicalskills and deepen our understanding of machine [13]. T.Ruotsalo,J.Peltonen,M.J.Eugster,D.Głowacka,
learning and user behavior analytics. P.Floreen,P.Myllymaki,G.Jacucci,
andS.Kaski,‘‘Interactiveintentmodelingfor
REFERENCES exploratory search,’’ ACM Trans. Inf. Syst., vol.
36,no.4, p.44,Oct.2018,doi:10.1145/3231593.
[1]. L.S.Vygotsky,MindinSociety:TheDevelopment of [14]. S.K.Shivakumar,‘‘Asurveyandtaxonomy ofintent-
Higher Psychological Processes. Cambridge,M based code search,’’ Int. J. Softw. Innov., vol.9,no.1,
A,USA:HarvardUniv.Press,1978. pp.69–110,Jan. 2021.
[2]. K.A.Mills,‘‘Shrekmeetsvygotsky:Rethinking [15]. P.Ren,Z.Liu,X.Song,H.Tian,Z.Chen,Z.Ren, and M.
adolescents’ multimodal literacy practices in de Rijke, ‘‘Wizard of search engine:
schools,’’J.AdolescentAdultLiteracy,vol.54,no.1, Accesstoinformationthroughconversationswith search
pp.35–45,Sep.2010. engines,’’ in Proc. 44th Int. ACM SIGIR Conf.Res.
[3]. A. Halevy, C. Canton-Ferrer, H. Ma, U. Ozertem, P. Develop.Inf.Retr.,Jul.2021.
Pantel, M. Saeidi, F. Silvestri, and V. Stoyanov,‘‘
Preservingintegrityinonlinesocial networks,’’
Commun. ACM, vol. 65, no. 2, pp. 92–98,Feb.2022.
[4]. X.Feng, X. Wang, and Y. Zhang, ‘‘Research on the
effect evaluation and the time-series evolution
ofpublic culture’s Internet communication under the
background of new media: Taking theinformation
disseminationofred tourism culture as an example,’’
J. Comput. Cultural Heritage,vol.16,no.1,pp.1–
15,Mar.2023.
[5]. C. I. Eke, A. A. Norman, L. Shuib, and H. F. Nweke,
‘‘A survey of user profiling: State-of-the art,challe
nges,andsolutions,’’IEEEAccess,vol.7, pp.144907–
144924,2019.
[6]. J. Liu, M. Mitsui, N. J. Belkin, and C. Shah, ‘‘Task,
information seeking intentions, and user behavior:
Toward a multi-level understanding of web search,’’
in Proc. Conf. Human Inf. Interact. Retr.,Glasgo
w,U.K.,Mar.2019,pp.123–132.
[7]. J. L. Hale, B.J.Householder, andK. L. Greene, ‘‘The
theory of reasoned action,’’ in The Persuasion
Handbook:Developmentsin Theory and Practice, vol.
14. Newbury Park, CA, USA: Sage,2002,pp.259–
286.
[8]. J. Shi, P. Hu, K. K. Lai, and G. Chen, ‘‘Determinants
of users’ information dissemination behavioron
socialnetworkingsites:Anelaboration likelihood
model perspective,’’ Internet Res., vol.28,no.
2,pp.393–418,Apr.2018.
[9]. P.Bedi,S.B.Goyal,A.S.Rajawat,R.N.Shaw,and
A.Ghosh,‘‘Aframeworkfor personalizing atypical
web search sessions with concept-based user profiles
using selective machine learning techniques,’’ in
Advanced Computing and Intelligent Technologies
(Lecture Notes in Networks andSystems),vol.21
8.Singapore:Springe,2022.52

IJISRT25APR1128 www.ijisrt.com 2252