An IoT Assisted Weather Prediction and

2023 International Conference on Advances in Computing, Communication and Applied Informatics (ACCAI) | 979-8-3503-1590-5/23/$31.00 ©2023 IEEE | DOI: 10.1109/ACCAI58221.2023.10200067

Information Monitoring Scheme based on

Intensive Learning Strategy
Sonia Jenifer Rayen, Minu Susan Jacob, S.Nivetha, Lekshmi S Raveendran, J. SarojiniPremalatha, S. Scinthia Clarinda

Assistant Professor, Department of computer science and Engineering, School of computing, Sathyabama Institute of Science and
Technology, Chennai – 600119

E-mail : [email protected], [email protected], [email protected],

[email protected], [email protected], [email protected]

Abstract- Several of the environmental characteristics everyday living cannot be overstated. Agriculture,
we hold dear, including our fundamental ecology, manufacturing, construction, and other sectors all
regional climatic variances, and global variety, are face difficulties as a result of environmental
threatened caused by widespread industry and conditions. However, the effects have been observed
increasing urbanization. Problems arise in our
mostly in commercial and agricultural settings. We
everyday lives as a consequence of the challenges we
face due of the rapid pace at which things are
are all aware of the significant importance that
changing. These problems have their origins in the agriculture plays in India's economy. Almost a
rapid pace of digitalization and the lack of enough quarter of India's GDP comes from the agricultural
infrastructure for handling and analyzing extremely sector. Smart agriculture has been a much discussed
large data sets. When data generated by the IoT issue in recent years. The smart term in IoT refers to
detection layer is inaccurate, partial, or irrelevant, it the improved performance achieved by reducing the
affects weather forecast reports, which in turn number of input variables. Land, water, and time are
disrupts activities that rely on precise and reliable all conserved, while modern agricultural
weather predictions. Some of the main problems in the technologies and scientific knowledge are put to
modern world is getting accurate readings of the
better use [1]. There are several established security
surroundings in real time, as doing so presents a
number of obstacles. The proposed technology could
methods for the Internet of Things (IoT), but there
avoid such events from occurring by constantly are also many potential ways to further strengthen it.
monitoring the weather. The suggested solution is
built on the cutting edge of embedded system These days, IoT implementations may be found in
technology and the IoT. Electronic components, sensor every facet of daily life [6]. Inventions in IoT aim to
nodes, and wireless infrastructure are also part of the manage and track routine activities, environmental
system. The primary function of the sensors in this conditions, and animal behavior [2]. Consequently,
framework is to monitor environmental factors such these improvements are made to make human labor
as temperature, humidity, and gas concentrations. easier and daily living more convenient. At this
Data collected in this way can be transmitted to
point, atmospheric monitoring techniques are
external systems, such as databases. Random Forest,
Logistic Regression, AdaBoost, and Naive Bayes were
developed to help keep tabs on and regulate things
all used to compare their respective identification of like temperature, humidity, and carbon dioxide
anomalies scores. These algorithms were applied to the levels.
time-based, temperature-based, pressure-based, and
humidity-based data produced by the sensors. Humans have spent a lot of time trying to figure out
their surroundings. Therefore, people have
Index Terms—Machine learning; Random Forest; developed several tools to gauge a wide range of
Logistic Regression; AdaBoost; Naive Bayes; Weather quantities. The thermometer, barometer, and
Prediction.
pyrometer are all instruments developed by humans
to measure different aspects of the environment.
I.INTRODUCTION
Traditional methods, nevertheless, may only be
employed domestically. But now, thanks to IoT, we
The importance of keeping tabs on the weather in
can monitor everything about our surroundings from

Authorized licensed use limited to: BRAC UNIVERSITY. Downloaded on March 24,2025 at 09:43:46 UTC from IEEE Xplore. Restrictions apply.
afar [3]. This means that determining environmental threat to life on Earth as well as one of the most
characteristics and quantifying them is much easier pressing ecological and socioeconomic problems of
now than it was in the past. the twenty-first century. Global warming has already
exceeded the threshold of what creatures and plants
Detectors including temperature, humidity, and can endure, leading to the extinction of many
pressure sensors form the basis of modern species, including coral reefs and forests. Extreme
atmospheric surveillance systems. Many of these weather has put millions of people at danger of
devices are adaptable to various settings. However, starvation and water shortages. In order to lessen
there are those who have prerequisites. These types these and comparable negative effects and
of detectors can measure the relevant chemical or implement efficient interventions, the ability to
physical weather data and translate them into an create precise and trustworthy estimations is crucial.
electrical output. As a result, the data is sent to a
digital card in the form of an electric signal. This last The suggested technique utilizes the data pre-
component may interpret the incoming signals and processing approach prior to sending information to
assign a value to every one of them. the server. At first, the layer that senses is used to
gather raw data from the objects. Once the raw data
There has to be an item that has sensors, actuators, was collected, classifiers were employed to check
and a way to communicate in order to create an IoT for data contamination. Sending pre-processed data
system [4]. The device must have a means of to the server for analysis reduces the amount of labor
connecting to the internet so that data may be sent to required to analyze the data. The outcomes of our
the cloud or a distant system for analysis and replicated model for IoT-based prediction of the
surveillance. Also, the device can behave in weather were analyzed. When compared to research
response to analytics-driven control data. There are on prediction of weather and changes in the climate,
several steps involved in gardening before the identifying anomalies across the IoT borders, the
harvest, and the weather is a major factor in all of recognition layer, as well as the communication
them. Because of the region's steep terrain, rain is a layer makes the data provided by the sensors more
common occurrence in Gorakhpur. Heavy rain is the useful and controllable. Furthermore, by doing so,
root of the farming crisis in this region. The ability service providers, consumers, and various other
to predict the weather is crucial in this context, both participants will experience less of the negative
for planning purposes and for maximizing consequences of anomaly data. Analytics is
agricultural yields. Therefore, in this case, farmers shorthand for the underlying machine learning
might benefit from a system for weather technique, in this case the model known as logistic
surveillance [5] by keeping an eye on the state of the regression. This model is educated using historical
atmosphere. If the quantity of water were to readings of climate, humidity, and luminosity. It's
increase, it would create problems for farmers. also put to use in the art of prediction.
Having enough water is essential while planting
paddy, but an increase in water supply might present II.RELATED STUDY
difficulties for farmers. If the weather's condition
can be predicted, it would be helpful in alleviating The Real-Time Weather Forecasting System is an
this issue. The primary goal of our method is to offer advanced tool for monitoring the current climate in
a signal as to whether or not it would be helpful for real time. Present climate analysis might be used to
farmers to do so before planting or harvesting. make forecasts about the atmosphere in the future.
Author [7] depicts and describes an effective
In response to both natural and human-caused prediction system that makes use of data gathered
processes, the land cover of Earth is constantly from sensors saved in the cloud to make remote
evolving. Natural processes including tsunamis, variable forecasts and trend projections over the
erosion, landslides, and floods are just a few internet. The suggested framework will illustrate the
examples of how land cover may change. Among intelligent system that makes use of WiFi for data
the most important factors in human-caused land transfer. Complex gear is employed in today's
use/cover alteration is urbanization. Constant standard prediction of the weather systems. Periodic
alterations to land use at various spatial scales have variations are prevalent in machinery because of
the potential to have both localized and systemic numerous moving components that require periodic
impacts on the weather. Climate change is an urgent

Authorized licensed use limited to: BRAC UNIVERSITY. Downloaded on March 24,2025 at 09:43:46 UTC from IEEE Xplore. Restrictions apply.
repair. Heavy and large-sized gear makes installation due to its ever-changing characteristics, accurate
in rural places without sufficient space extremely weather predictions may be challenging. In this [9],
challenging, if not impossible. The existing system's the author presents a method for forecasting the
biggest flaw is that it constantly has issues, such its weather in Sedati, a subdistrict of Sidoarjo Regency,
inability to alert regulators to impending poor Indonesia, on every day. In order to predict future
weather or its inability to predict rapid changes in weather, the system employs the Gated Recurrent
the weather. This is due to the fact that cables are Unit (GRU) technique, which is supplied with
used to manually transfer data obtained by various information from the BMKG Juanda Meteorological
equipment. Weather damage and human casualties Station's archives. The GRU regression equation
are only two examples of the kinds of problems that achieved the lowest median RMSE on the
might result from this kind of delay. As the world experiment's information pertaining to forecasting
around us becomes more technologically advanced, conditions across four days, including relative
so do more things. Our suggested system is an IOT- humidity, wind speed, temperature, sea level
based smart prediction of the weather technology pressure, maximum wind speed, minimum
that aims to be both easy to adopt and economical. temperature, maximum temperature, and dew point.
This tiny construction will make it possible to set up After being coupled with IoT devices, the reliability
weather stations in otherwise inaccessible regions. of the sequential GRU model for rain categorization
The detectors we're using in this study are also more dropped from 0.88 to 0.75.
cost-effective than those employed by conventional
weather prediction models. Therefore, we develop a The writer of article [10] suggests a cloud-based,
project that minimizes expenses while making intelligent weather station. Raspberry Pi is used in
reliable predictions. Most regions of a country like this system to gather and analyze weather data. To
India, which rely heavily on agriculture, must have foresee the consequences of this weather shift, the
access to an accurate weather forecasting system. collected data is stored and processed on the cloud.
The equipment is built to accurately track and report
The Internet of Things (IoT) and Machine Learning on meteorological parameters such air temperature,
(ML) are now the center of attention. There is a wide humidity, wind speed, barometric pressure,
variety of IoT hardware platforms to choose from. precipitation, and more. The goal is to create a
The primary objective of these chips is the ESP8266. system that can be implemented with little money,
In article [8], a researcher develops a real-time effort, and time spent on it. The Raspberry Pi, a
weather forecast system applicable to a wide range variety of sensors, and Wi-Fi all contribute to the
of settings, including private residences, commercial method's low power consumption and inexpensive
establishments, agricultural operations, public cost of construction. The data from the sensors is
venues, and sports arenas, among others. The system collected by Raspberry Pi Zero W boards and sent to
makes use of a light intensity sensor (LDR) as well a larger Raspberry Pi 3 board at the base station. The
as a temperature and humidity sensor (DHT11). information is then sent from the Raspberry Pi 3
NodeMCU and an ESP8266-01 module are used to over Wi-Fi to a cloud database, where it is utilized to
transmit the information gathered from the detectors build a new Machine Learning model that may be
to a ThingSpeak cloud server. A custom HTML used to anticipate the impact and investigate weather
website is also provided for viewing the information behaviors and trends. The Django-based software is
in real time. The artificial intelligence ecosystem is hosted in the cloud, allowing users to access weather
established using a model based on logistic information and insights from anywhere, at any
regression. Pre-recorded data collected from sensors time.
is used for training this model. In addition,
NodeMCU logs information from sensors like In article [11], the writer presents the results of an
humidity, temperature, and light intensity, and then investigation into the feasibility of using machine
transfers those numbers to a Jupyter notebook learning algorithms for regression to forecast the
running in a python framework. The model is put to weather in agricultural areas in preparation for
the test, and predictions are made for a given value, automated irrigation systems. The suggested
all in real time, by flashing the light attached to technique takes into account both the time of day as
NodeMCU. well as the location of the field to provide accurate
forecasts of temperature, precipitation, wind, and
The environment affects everybody in some way, yet evapotranspiration. Several machine learning

Authorized licensed use limited to: BRAC UNIVERSITY. Downloaded on March 24,2025 at 09:43:46 UTC from IEEE Xplore. Restrictions apply.
methods were tried out, with the results compared to
Humidity Sensor
find the optimum model for doing this. These
methods included Linear Regression, Decision Tree,
Random Forest, and Neural Networks. Cross-
DS1302
validation results favor Random Forests and
Decision Trees for their efficacy. The approach, its
execution, and the experimental findings are all
Power Suuply Atmega328 ESP32
described in great depth in this publication.

III.METHODOLOGY
MQ-7

In order to improve the precision and consistency of

weather forecasts, this study concentrates on
MQ-131
recognizing anomalies on a stream of information
comprised of four-dimensional temperature, Buzzer

humidity, pressure, and time information. During

anomaly detecting, a mock version of the Figure 1. The system's block layout.
information stream from the IoT sensor layer is
used. Among the first and second levels of the Dataset
canonical IoT, anomaly detection is carried out in a In all, 9690 points of information were generated by
simulated environment. By complementing the ideas the study's IoT sensor layer sensors, which included
of precision and dependability, this pretreatment four-dimensional temperature, humidity, pressure,
layer that is going to focus on the role of edge and time readings. Approximately 6000 records
computing, aids in the production of more insightful (75% of the total) are employed for training,
big data. whereas 2900 records (30% of the total) have been
utilized for confirmation and forecasting. The scikit-
The building blocks of an IoT system are the raw learn library is also used for modelling and
data gathered by IoT gadgets. Most useful IoT normalization operations.
services are created through the collection and
processing of data from objects. Data collection The information was analyzed using an exploratory
makes use of a broad variety of wireless equipment, method. The investigation covered a wide range of
such as wireless sensor networks (WSN), low power factors, including the ones related to time,
wide area networks (LPWAN), Wi-Fi, Bluetooth, temperature, humidity, pressure, and pressure.
and mobile phone networks. Information in a Features with a strong influence on the outcome
network is managed by the server. The information were incorporated in the conditioning set once the
generated by IoT gadgets is uploaded to a remote association matrix was constructed.
server. The server does a search on the collected data
to find what it needs. Obviously, vital information is An Arduino cart, temperature, pressure, and
transmitted to smart IoT services. When analyzing humidity detectors, as well as a time supervisor,
large amounts of data, its reliability becomes were utilized to generate the raw data needed in the
increasingly important. The confidentiality of anomaly identification process. Temperature,
information reduces the server's computational load pressure, and humidity readings were taken with the
during data analysis from an Internet of Things (IoT) use of an Arduino UNO R3 and a BME280 sensor
device. The server's processing burden drops as data throughout the development of the base.
is handled, which is good for the environment. Furthermore, the exhibited DS1302 clock module
Consequently, the integrity of data must be was employed to record and store real-time data
supported by the system. The schematic production information. The ESP32 microcontroller
representation of the system is shown in Figure 1. was utilized to record and transmit the collected data
in its raw form.

Authorized licensed use limited to: BRAC UNIVERSITY. Downloaded on March 24,2025 at 09:43:46 UTC from IEEE Xplore. Restrictions apply.
Arduino UNO R3 manufacturing calibration can be reduced or
The Figure 2 Arduino UNO card has been eliminated when sensors are interchangeable. There
implemented in our system. It seems to be a is information accessible for every sensor's
microprocessor integrated inside a tiny electronic calibration. The HIH-4000 Series provides RH
card. The Arduino UNO card may be configured to (Relative Humidity) detection capabilities at par
read data from sensors and operate motors and other with that of instruments at an affordable price and in
mechanical devices. However, Arduino's digital and a solder able SIP.
analogue pins allow for connections to be made to
the detectors and actuator. The Arduino may also be DS1302 Module
powered by a Direct Current (DC) port or a USB For example, as can be seen in Figure 4, the DS1302
connection. An inbuilt voltage regulator is connected trickle-charge timekeeping chip incorporates both a
to an Arduino UNO card. Therefore, the connected real-time clock/calendar as well as 31 bytes of static
power supply must be between 5 and 12V. RAM. Through a basic serial interface, it can talk to
a CPU. Time, date, month, and year are all
displayed in the real-time clock/calendar. The
clock/RAM may be accessed using just three wires
(called CE, I/O, and SCLK, respectively) for
communication. One byte at a time or in bursts of up
to 31 bytes can be sent and received between the
host computer and the clock/RAM. The DS1302
requires less than 1W to run and can store data and a
clock with precision. The DS1302 features two
power connectors, one for the main supply and one
as a secondary.
Figure 2. Arduino UNO R3.

HIH-4000 Humidity Sensor

Figure 4. DS1302 Real Time Clock Module.

ESP32 Microcontroller

Figure 3. HIH-4000 Humidity Sensor

OEM (Original Equipment Manufacturer) Figure 5. ESP32 Microcontroller.

manufacturers can benefit most from the HIH-4000
Series Humidity Sensors depicted in Figure 3. This As can be seen in Figure 5, Espressif's ESP32 family
sensor's near linear voltage output allows it to be of low-cost, low-power System-on-a-Chip (SoC)
used as a direct connection to a controller or other microcontrollers features both Wi-Fi and Bluetooth
device. The expenses associated with OEM wireless connectivity and a dual-core CPU. To

Authorized licensed use limited to: BRAC UNIVERSITY. Downloaded on March 24,2025 at 09:43:46 UTC from IEEE Xplore. Restrictions apply.
further conserve energy, the ESP32 enables low- to a CPU. Seconds, minutes, hours, days, dates,
power operating conditions like deep sleep in months, and years may all be seen on a real-time
addition to its already low power consumption as in clock or calendar? It was crucial in collecting
comparison to other microcontrollers. For Internet of reliable measurements of temperature, pressure, and
Things (IoT) and Home Automation (HA) humidity for the study. Another part of the project is
assignments, the ESP32's ability to link to a Wi-Fi the ESP32, a multipurpose microcontroller with
network to browse the World Wide Web (station built-in Wi-Fi and Bluetooth connectivity. IoT
mode) or establish its own Wi-Fi wireless application development is facilitated by its
connection (access point mode) and allow other structure. The ESP32 is reliable even in harsh
gadgets to link to it is crucial. environments, since it can operate between -40 and
+125 degrees Celsius.
Sensors
Figure 6 depicts the variety of detectors in our The kinesis stream of information will be used, as it
system's architecture. Those gadgets include receives information in real time and processes it
thermometers, humidity monitors, and carbon alongside the kinesis analysis of information. The
dioxide detectors. Temperature, ozone (O3/level), data is subsequently written to the specified location.
Sulphur dioxide (SO2/level), carbon monoxide The temperature sensor is an example of an IoT
(CO/level), nitrogen monoxide (NO2/level), and gadget since it can mimic data movement across the
relative humidity are only some of the air internet. The calculated temperatures have been used
characteristics that will be measured by these as the norm. Anomaly temperature readings are
detectors. The sensor at issue provides the those that go beyond the usual and are subsequently
straightforward power that is associated with a filtered out. A stream of information is formed by
selected climatic component. The microprocessor in combining the produced data with Google Colab
the Arduino will then use this voltage to calculate a services. Whenever the unusual information has
value. been found, the reliable and correct data is isolated
from the anomaly and sent to its final destination in
order to preserve data integrity.

IV.RESULTS AND DISCUSSIONS

We have conducted experimental experiments in an

effort to gauge the efficacy of our suggested method.
The source code was compiled, and then uploaded to
an Arduino board, where we ran tests on each
Figure 6. MQ-7 and MQ-131 variable.

The Atmega328-based Arduino Uno microcontroller Classification Reports

board features a USB connection, power jack, ICSP Precision, recall, accuracy, and F1-score are only
header, reset button, 14 digital I/O pins (6 of which few of the criteria used to assess the results of this
may be used as PWM outputs), and 16 MHz investigation. The confusion matrix is depicted
oscillator. The HIH-4000 was utilized since it is a graphically for your convenience. The confusion
digital sensor that measures temperature, pressure, matrix has a dimension of 2 by 2. The matrix has
and humidity all in one. It uses well-established four values, and they are (in order) TP, TN, FP, and
concepts of detecting. The accuracy of its FN. The sensitivity, specificity, accuracy, and error
measurements for temperature and pressure are 1 °C rate are only few of the metrics that can be
and 1 hPa, respectively, while for humidity it is 3%. calculated with the use of the confusion matrix.

The DS1302 is a high-quality, dual-power-supply It has therefore been attempted to use categorization
calendar/real-time clock microcontroller. It operates success as well as comparability matrices to
smoothly over a wide 2.5 V to 5.5 V supply voltage establish which method performs the best for
range. There's also 31 bytes of non-volatile memory preparing information in predictions of the weather.
on the chip. Using a basic serial interface, it can talk

Authorized licensed use limited to: BRAC UNIVERSITY. Downloaded on March 24,2025 at 09:43:46 UTC from IEEE Xplore. Restrictions apply.
 1400

1200

1000

CO Concentration
800

600

400

200

0
5 10 15 20 25 30 35 40 45 50 55 60 65

Time (s)

Figure 9. CO air concentration experiment outcome

Table 1. Findings from the monitoring of pollution

Figure 7. Correlation Matrix. levels in the environment.
Sit Temperatu Humidi Pollutant concentration
Figure 7 shows a direct relationship among pressure, e re (0C) ty (%) (x10-6)
humidity, and temperature in the form of a O3 SO2 CO NO2
correlation matrix. The relationship among time and 1 25 66 0.04 63 1246 0.8
temperature displays a similar ratio. The correlation
2 29 61 0.05 73 1325 2.1
among the variables shows that time and
temperature readings are superior at detecting 3 23 72 0.04 66 1421 1.3
outliers. We use the impacts of temperature,
humidity, and pressure as a training set as these
factors are highly correlated. 80 1450

70
1400
The outcomes of these experiments on SO2 60

concentration in air are depicted in Figure 8. Figure 50

1350

9 depicts the CO concentration test result. The 40 1300

sensor's wires have been lengthened so that it may 30
1250
be installed in a compact enclosure that is then 20
connected to the outside air. A compound reaction 10
1200

takes place at a measuring instrument, and CO gas is 0 1150

released into the container as a byproduct. To speed 1 2 3
up the reaction time of the chemical within, this is
being set on a heater. Temperature (0C) Humidity (%) O3

SO2 NO2 CO

80 Figure 10. Location-specific measurements of air

70 pollution levels.
60
SO2 Concentration

50 The simulated installation's findings for collecting

40 sub-Application Programming Interface (API)
30 numbers in low-traffic as well as high-traffic
20 locations are shown in Table 1 and Figure 10. The
10
test site was set up on the side of the road in a
0
residential neighborhood close to the manufacturing
5 10 15 20 25 30 35 40 45 50 55 60 65
hub. For a single day, we took 30 observations at
Time (s)
regular intervals. All of the data was collected and
stored in the system.
Figure 8. The levels of SO2 in the air were measured
in an experiment.

Authorized licensed use limited to: BRAC UNIVERSITY. Downloaded on March 24,2025 at 09:43:46 UTC from IEEE Xplore. Restrictions apply.
Our goal in developing this "weather monitoring 1
system using IOT" was to create a tool for tracking 0.99
the state of the atmosphere and collecting data on 0.98
external variables in real time. A parameterized 0.97
environmental surveillance system prototype is 0.96
currently under construction. Multiple weather 0.95
variables are tracked by the system that has been 0.94
built. The information seen is being saved to the web 0.93
server by means of HTTP requests made to the page. Precision Recall F1-Score

Table 1 displays the combination of parameters of Figure 12. Logistic regression classification report.
the information utilized for confirmation and
forecasting and the influence of the RF technique on Finding the best settings required a grid search
categorization. Grid search has been employed with driven by the LR technique. Both the C and
the random forest method to get the optimal punishment variables for the categorization criterion
parameters. There are 200 maximum estimators, 4 were settled upon as 0.1 and l2, respectively.
maximum depth levels, and an auto maximum
features level. The requirement variable is calculated Figure 13 displays the Adaboost technique's
using gini within the constraints of the categorization outcome. Metrics were affected by
categorization criteria. The report of the random the verification as well as forecasting data, which
forest categorization is displayed in Figure 11. accounted for 30% of the total. Grid-search was
applied to the Adaboost technique to find the
Table 1.Analysis of Model Classification. optimal settings. The variables for the base
Model Precision Recall F1-Score estimating criterion are gini, the base splitter
Random Forest 0.99 0.98 0.96 requirement is best, as well as the n_estimators
Logistic requirement is set to 1.
0.98 0.99 0.95
Regression
AdaBoost 0.97 0.95 0.98
0.985
Naive Bayes 0.95 0.98 0.99
0.98
0.975

0.995 0.97

0.99 0.965
0.985 0.96
0.98 0.955
0.975 0.95
0.97 0.945
0.965 0.94
0.96
0.935
0.955
Precision Recall F1-Score
0.95
0.945
Precision Recall F1-Score Figure 13. Adaboost classification report.

Figure 11. Random forest classification report. The NB technique's categorization output is
displayed in Figure 14. The results demonstrate the
Figure 12 shows the categorization result from the effect of the verification and forecasting
LR technique. Findings reveal how data comprising information, which is 30% of the whole data set. The
30% of the set influences the measures. grid-search strategy was chosen over the NB
technique for determining the optimal variables.
Considering the bounds of the categorization
criteria, the value of the var_smoothing variables
was determined to be 1.0.

Authorized licensed use limited to: BRAC UNIVERSITY. Downloaded on March 24,2025 at 09:43:46 UTC from IEEE Xplore. Restrictions apply.
 1
occurs among the sensor and network layers. The
investigation compares the abilities of five
0.99
techniques at filtering data by conducting the
0.98
identification of anomalies on the data flow
0.97 produced among the identification layer and the
0.96 network layer. Techniques may be tested and refined
0.95
with the help of Google Colab's simulation tools.
The experimental findings portion of the study
0.94
contains comparisons between and simulation
0.93
outcomes. The experimental findings portion of the
Precision Recall F1-Score
study contains the comparison and simulation
Figure 14. Naive Bayes classification report. outcomes. SVC, Adaboost, LR, NB, and RF
techniques were instantly applied to the data stream
Table 2 and Figure 15 display the running time in to classify it. Techniques allow for real-time
seconds required to process data in RF, LR, categorization of incoming data before it reaches the
Adaboost, NB, and SVC. destination. We measure how well an anomaly
detector works by its sensitivity, recall, and f1 score.
Table 2. Comparison of data processing speed Separating out anomalous information from the
mountain of data generated by the IoT detection
Execution Time layer is an integral part of this real-time process. The
Logistic Regression 2.19s speed with which various algorithms process data is
Naive Bayes 1.92s also compared.
Support Vector Machine
46.48s If we take a look at the time required for execution
Classification
of the RF technique, for instance, we can see that it
Random Forest 52.83s
performs poorly in our tests. As indicated in the
Adaboost 2.20s introduction, this circumstance is crucial for
identifying the best techniques for producing the
60
most reliable weather prediction. Test outcomes and
50
the study's central IoT-based climate prediction
system demonstrate the importance of removing
Execution Time

40

30 deviations from recognizing layer data using a

20 suitable data preprocessing strategy for achieving
10 superior outcomes from IoT-based systems.
0
Logistic Naive Bayes Support Random Adaboost V.CONCLUSION AND FUTURE SCOPE
Regression Vector Forest
Machine
Classification
Our suggested system includes low-cost weather
Models
monitoring devices. Client-side architecture is the
basis for the suggested system's operation. The
Figure 15. Comparison of data processing speed suggested method utilized multiple detectors to
monitor a wide range of surrounding data. The
This research examines how the Internet of Things developed system requires fewer sensors than the
(IoT) may be used in meteorology. This may have baseline version. Our suggested model's primary
seen, weather prediction systems employ the objective is to reduce the system's overall
standard four-layer IoT structure consisting of operational cost. For anyone's unrestricted use. Our
identification, network, service, and application. The suggested system would collect information from a
suggested data pre-processing methodology aims to number of different sensors and upload the results to
improve the prediction of weather IoT research by a central server. There have been preliminary tests of
mitigating problems that might develop in systems a system to check air and weather characteristics by
constructed using the conventional architecture. The utilizing the IoT for tracking environmental factors.
proposed organizational model differs from the The technology offers an efficient means of creating
standard IoT architecture in that data pre-processing

Authorized licensed use limited to: BRAC UNIVERSITY. Downloaded on March 24,2025 at 09:43:46 UTC from IEEE Xplore. Restrictions apply.
a weather station with little power usage. The [5] Kondaka, L.S., Thenmozhi, M., Vijayakumar, K. et al. An
intensive healthcare monitoring paradigm by using IoT based
technology has been effectively tested in a machine learning strategies. Multimed Tools Appl (2021).
controlled indoor setting, where it has updated the https://doi.org/10.1007/s11042-021-11111-8.
ambient and weather conditions based on data [6] S. Deepika and K. P. Vijayakumar, "IoT based Elderly
Monitoring System," 2022 6th International Conference on
gathered from sensors. Random Forest, Logistic Trends in Electronics and Informatics (ICOEI), 2022, pp. 573-
Regression, AdaBoost, and Naive Bayes were all 579, doi: 10.1109/ICOEI53556.2022.9777186.
[7] Shivanshu, P. Nagwanshi and A. Chauhan, "Smart Real Time
used to compare their respective identification of Weather Forecasting System," 2021 3rd International
anomalies scores. Random Forest was the most Conference on Advances in Computing, Communication
accurate of these machine learning algorithms. Control and Networking (ICAC3N), Greater Noida, India, 2021,
pp. 558-562, doi: 10.1109/ICAC3N53548.2021.9725697.
Additionally, the system may be adapted for [8] G. Verma, P. Mittal and S. Farheen, "Real Time Weather
commercial usage, making it useful in various Prediction System Using IOT and Machine Learning," 2020 6th
contexts like smart homes, buildings, sports, International Conference on Signal Processing and
Communication (ICSC), Noida, India, 2020, pp. 322-324, doi:
hospitals, etc. 10.1109/ICSC48311.2020.9182766.
[9] H. Darmawan, M. Yuliana and M. Z. SamsonoHadi, "Realtime
Weather Prediction System Using GRU with Daily Surface
REFERENCES Observation Data from IoT Sensors," 2022 International
Electronics Symposium (IES), Surabaya, Indonesia, 2022, pp.
[1] M. Sadhukhan, S. Dasgupta and I. Bhattacharya, "An Intelligent 221-226, doi: 10.1109/IES55876.2022.9888468.
Weather Prediction System Based on IOT," 2021 Devices for [10] P. Kapoor and F. A. Barbhuiya, "Cloud Based Weather Station
Integrated Circuit (DevIC), Kalyani, India, 2021, pp. 528-532, using IoT Devices," TENCON 2019 - 2019 IEEE Region 10
doi: 10.1109/DevIC50843.2021.9455883. Conference (TENCON), Kochi, India, 2019, pp. 2357-2362,
[2] R. Pavaiyarkarasi, et al "A Productive Feature Selection doi: 10.1109/TENCON.2019.8929528.
Criterion for Bot-IoT Recognition based on Random Forest [11] F. Raimundo, A. Glória and P. Sebastião, "Prediction of
Algorithm," 2022 IEEE 11th International Conference on Weather Forecast for Smart Agriculture supported by Machine
Communication Systems and Network Technologies (CSNT), Learning," 2021 IEEE World AI IoT Congress (AIIoT), Seattle,
2022, pp. 539-545, doi: 10.1109/CSNT54456.2022.9787583. WA, USA, 2021, pp. 0160-0164, doi:
[3] L. Varghese, G. Deepak and A. Santhanavijayan, "An IoT 10.1109/AIIoT52608.2021.9454184.
Analytics Approach for Weather Forecasting using Raspberry Pi
3 Model B+," 2019 Fifteenth International Conference on
Information Processing (ICINPRO), Bengaluru, India, 2019, pp.
1-5, doi: 10.1109/ICInPro47689.2019.9092107.
[4] Sugadev, et al (2022). Implementation of Combined Machine
Learning with the Big Data Model in IoMT Systems for the
Prediction of Network Resource Consumption and Improving
the Data Delivery. Computational Intelligence and
Neuroscience, 2022.

10

Authorized licensed use limited to: BRAC UNIVERSITY. Downloaded on March 24,2025 at 09:43:46 UTC from IEEE Xplore. Restrictions apply.