Breath of the Future: Predicting Air Quality Index with ML and IoT

Sneha Varur

, Uma M Hiremath

, Devaraj Hireraddi

, Nitin Nagaral

, Kushalagouda Patil

and Gouri Vernekar

KLE Technological University, Hubballi, India

Computer Science and Engineering, KLE Technological University, Hubballi, India

Keywords:

Air Quality Indicators, Machine Learning, Internet of Things, Predictive Solutions, Artiﬁcial Neural Net-

works, Environmental Monitoring, Root Mean Square Error, Sensor Integration, IoT Infrastructure, Public

Health Protection.

Abstract:

Air pollution continues to pose substantial threats to both public health and environmental stability, emphasiz-

ing the critical need for sophisticated monitoring and predictive solutions. The study offers a visionary strategy

for forecasting the Air Quality Index (AQI) by integrating Machine Learning (ML) with the Internet of Things

(IoT). The proposed system utilizes an array of environmental sensors to gather real-time data, which is sub-

sequently processed by advanced machine learning algorithms, with a speciﬁc emphasis on artiﬁcial neural

networks (ANN), to generate accurate AQI predictions. The IoT architecture facilitates seamless, real-time

data acquisition, enhancing both the accuracy and responsiveness of the system. This paper delves into the

technical aspects of the system, including the detailed methodology, hardware conﬁguration, and software

integration, to illustrate the synergistic potential of ML and IoT in air quality forecasting. The results indicate

strong model efﬁcacy, with a root mean square error (RMSE) of 82.84% and a classiﬁcation accurateness

of 94.54%, underscoring the system’s capability in effectively monitoring and predicting air pollution levels.

The research offers signiﬁcant advancements in the ﬁeld of environmental monitoring, demonstrating how the

convergence of ML and IoT can play a pivotal role in the future of air quality management and public health

protection.

1 INTRODUCTION

Atmospheric corrosion remains a major global con-

cern, with substantial risks to ecosystems, human

health, and climate stability. With the swift pace of in-

dustrialization and urbanization in contemporary so-

ciety, the concentration of harmful pollutants in the

atmosphere has reached unprecedented levels. Pollu-

tants such as nitrogen dioxide (NO2), sulfur dioxide

(SO2), carbon monoxide (CO), ozone (O3), ammonia

(NH3), and particulate matter (PM 2.5 and PM 10)

(Rakib, Haq, et al. 2022) have been directly linked to

respiratory and cardiovascular diseases, cancer, and

various neurological disorders. The urgency to mon-

itor and manage air quality is more critical than ever,

especially in densely populated urban areas where the

adverse effects of pollution are most pronounced (Mi-

hirani, Yasakethu, et al. 2022). However, the com-

plexities involved in accurately measuring and pre-

dicting AQI make it a challenging task. One of the

primary challenges in air quality monitoring is the dy-

namic nature of pollutants, which vary signiﬁcantly

over time and space. Traditional methods rely on sta-

tionary monitoring stations, which, although accurate,

are often limited in their coverage and cannot provide

real-time data for every location of interest. This lim-

itation hampers the ability to respond swiftly to haz-

ardous pollution levels. Moreover, the vast amount

of data generated from multiple sources, including

weather conditions and trafﬁc patterns, complicates

the task of accurately predicting future pollution lev-

els. Effective monitoring systems must therefore be

capable of both real-time data acquisition and sophis-

ticated analysis to provide timely and accurate air

quality forecasts.

The smart device ecosystem has come to promi-

nence as a game-changing innovation with the po-

tential to change environmental surveillance. By en-

abling a network of interconnected sensors and de-

vices, IoT facilitates the continuous collection of real-

722

Var ur, S., Hiremath, U. M., Hireraddi, D., Nagaral, N., Patil, K. and Vernekar, G.

Breath of the Future: Predicting Air Quality Index with ML and IoT.

DOI: 10.5220/0013584300004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 1, pages 722-728

ISBN: 978-989-758-763-4

time data across large areas. In the context of air

quality monitoring, IoT devices can be strategically

deployed to gather data on various environmental pa-

rameters, including pollutant concentrations, temper-

ature, and humidity (Gupta, Mohta, et al. 2023).

This infrastructure signiﬁcantly enhances the ability

to monitor air quality at a granular level, providing the

necessary data to understand pollution patterns and

dynamics in real-time. However, the sheer volume of

data collected by IoT devices presents its own set of

challenges, notably on the subject of data processing,

storage, and analysis.

To address the challenge of analyzing large

datasets generated by IoT sensors, Machine Learn-

ing (ML) (M

endez, Merayo, et al. 2023) offers pow-

erful tools for making sense of complex and high-

dimensional data. ML algorithms excel at identify-

ing patterns and correlations that may not be imme-

diately apparent through traditional statistical meth-

ods. In the domain of air quality projection, ML mod-

els can be trained on historical data to forecast future

pollutant levels (Krishna and Nabi, 2022), bearing

into account various contributing aspects such as me-

teorological circumstances and trafﬁc data (Kumari,

Vasuki, et al. 2020). Among these models, Artiﬁ-

cial Neural Networks (ANNs) have shown particular

promise due to their ability to model non-linear rela-

tionships and learn from continuous streams of data

(Shaban, Kadri, et al. 2022). The integration of ML

with IoT not only enhances the accuracy of air quality

predictions but also enables adaptive learning, where

the model continuously improves as more data be-

comes available.

This study explores the convergence of IoT and

ML technologies to create an advanced air quality

monitoring and prediction system. By leveraging

real-time data collected from a network of environ-

mental sensors and applying ANN models, we have

developed a system capable of predicting AQI with

high accuracy. The system was tested using data

collected from sensors monitoring various pollutants

and environmental conditions. The experimental re-

sults demonstrated an RMSE of 82.84% and a pre-

cision of classiﬁcation outcomes of 94.54%, indicat-

ing the system’s effectiveness in predicting air quality.

These ﬁndings underscore the potential of combin-

ing IoT and ML technologies to address the pressing

challenge of air pollution and pave the way for more

responsive and informed environmental management

practices.

2 LITERATURE SURVEY

The study by the authors in (Gupta, Mohta, et al.

2023) attempts to predict the AQI in Indian cities

using Support vector regression, Random Forest Re-

gression (RFR), and CatBoost Regression (CR). They

incorporate the (SMOTE) Synthetic Minority Over-

sampling Technique for managing skewed datasets.

While their results indicate that RFR and CR perform

reasonably well, the improvements with SMOTE are

limited to speciﬁc cities. The overall approach lacks

a comprehensive evaluation across diverse environ-

ments, which limits its generalizability.

In (Bhattacharya and Shahnawaz, 2022), the au-

thors use Support Vector Regression (SVR) to fore-

cast air quality in New Delhi, achieving an accuracy

of 93.4%. The study highlights the signiﬁcance of

data pre-processing and demonstrates that operating

the full spectrum of variables yields more promising

results than feature selection via PCA. However, the

study is conﬁned to New Delhi and relies heavily on

archived data, which may not adequately represent

real-time prediction scenarios.

The research in (Gogineni and Murukonda, 2022)

compares multiple machine learning methods for

AQI prediction, including LASSO, SVR, and Ran-

dom Forest. While some methods, like Extra Trees

and Ridge Regression, showed promising results, the

overall performance was inconsistent across different

datasets. The study’s reliance on conventional regres-

sion models also limits its ability to handle complex,

real-world air quality dynamics effectively.

In (Murugan and Palanichamy, 2022), the authors

focus on predicting PM2.5 levels in Malaysian smart

cities using Random Forest and MLP (Multi-Layer

Perceptron). Though Random Forest achieved 97%

accuracy, the study’s scope is narrow, with ﬁndings

that may not translate well to other regions or pol-

lutants. The research lacks a detailed exploration of

how these models would perform under different en-

vironmental conditions or with varying data quality.

3 PROPOSED METHODOLOGY

The envisioned task entails designing and implement-

ing an air quality monitoring system, integrating mul-

tiple sensors with a microcontroller, and utilizing ma-

chine learning for predictive modeling. This section

outlines the experimental setup, sensor integration,

data processing techniques, and the selection and val-

idation of the prediction model. Additionally, a com-

parative analysis is furnished to demonstrate the con-

vincingness of the suggested guideline.

Breath of the Future: Predicting Air Quality Index with ML and IoT

723

3.1 System Overview

The system includes an Arduino Uno microcontroller,

an ESP8266 Wi-Fi module, and four sensors: a MQ-

135 for detecting ammonia (NH

), a MQ-7 for moni-

toring carbon monoxide (CO), a MQ-2 for further gas

detection, and a DHT11 for measuring temperature

and humidity. The detectors collect data in real-time,

which is then sent to the ThingSpeak cloud platform

for storage, analysis, and modeling. Figures 1 illus-

trate the hardware setup and 2 the system architecture.

Figure 1: Hardware Setup

The ﬁgure 1 illustrates a circuit diagram involving

an Arduino Uno microcontroller connected to multi-

ple sensors and a Wi-Fi module. Below is a detailed

breakdown of the components and their connections:

3.2 Components:

• Arduino Uno: The central microcontroller board

that controls and processes the data from the

probes.

• DHT Sensor: A digital humidity and temperature

sensor.

• MQ135 Sensor: A gas sensor used to measure air

quality (e.g., CO2, NH3, Benzene).

• MQ2 Sensor: A gas sensor that detects

ﬂammable gases like LPG, Propane, and Hydro-

gen.

• MQ7 Sensor: A gas sensor speciﬁcally designed

to detect carbon monoxide (CO).

• Wi-Fi Module (likely ESP8266): A module that

enables the Arduino to connect to a Wi-Fi network

for data transmission.

3.3 Wiring:

• DHT Sensor: Connected to a digital pin on the

Arduino (likely D4 or similar) for data input.

VCC (Power) - is linked 5Volt pin, and the GND

pin is concatenated to the ground connection of

the Arduino.

• MQ135, MQ2, and MQ7 Sensors: Each sensor

has 3 connections VCC is connected - 5V pin.

GND is connected - ground pin. The analog out-

puts are connected to different analog input pins

on the Arduino (e.g., A0, A1, A2).

• Wi-Fi Module: Attached to the Arduino’s TX

and RX pins for serial communication. VCC is

connected to the 3.3V pin (if it’s ESP8266) or 5V

pin (if it’s a different model). GND is connected

to the ground.

3.4 System Architecture

CO2 NO

CO2 NO2

NH3 SO2

Thing Speak

Cloud

ML Model

BUS Stand IoT System

Values of Gases

Data

Good

Moderate

Unhealthy

Unhealthy for

Strong people

Hazardous

AQI Bucket

OLED Display

Classification

Figure 2: System Architecture

The ﬁgure 2 represents a system designed to mon-

itor and assess air quality at a bus stand operating a

blend of IoT technology, cloud computing, and ma-

chine learning. The system begins by using sensors

to detect various gases like CO2, NO, NO2, NH3,

and SO2 in the environment. These gas concentra-

tion values are then transmitted to the ThingSpeak

cloud platform, where they are stored and processed.

The unprocessed data gathered from the sensors is fed

into a machine learning model, which analyzes the

data and classiﬁes the air quality into different vari-

eties, such as ”Good,” ”Moderate,” ”Unhealthy,” and

”Hazardous.” This classiﬁcation is based on the AQI.

Finally, the resulting air marker classiﬁcation is dis-

played on an OLED screen, providing real-time in-

formation about the air quality at the bus stand. The

system’s goal is to offer accurate, real-time air qual-

ity assessments, enabling people at the bus stand to

be aware of the pollution levels and make informed

decisions about their exposure.

INCOFT 2025 - International Conference on Futuristic Technology

724

3.5 Hardware and Software Integration

3.5.1 Sensor Integration

Each sensor is interfaced with the Arduino Uno mi-

crocontroller. The analog signals from the gas sensors

are converted to digital values using the Arduino’s

ADC. The DHT11 sensor provides digital readings

for temperature and humidity directly. The sensors

are connected as follows:

out

= R

− V

sensor

(1)

where V

out

is the output voltage, R

is the load re-

sistance, and V

sensor

is the sensor voltage.

3.5.2 Data Broadcasting

The ESP8266 wireless networking module is conﬁg-

ured to transmit sensor data to the ThingSpeak cloud.

Data is sent using HTTP POST requests, formatted as

JSON objects. The module operates in station mode,

connected to a local Wi-Fi network.

3.5.3 Data Preprocessing

The raw sensor data is preprocessed to handle incom-

plete data, outliers, and noise. Unrecorded entries are

ﬁlled in or estimated utilizing linear interpolation, and

outliers are detected and removed based on a z-score

threshold of 3. The data is then averaged on an hourly

basis to reduce temporal variability.

z-score =

− µ

(2)

where x

denotes the data values, µ represents the

mean, and σ signiﬁes the standard deviation.

3.6 Model Choosing and Training

3.6.1 Model Selection

The project explored diverse machine learning mod-

els for predictive analysis, including Linear Regres-

sion, polynomial regression, and LSTM webs. How-

ever, these models were either insufﬁcient or over-

complicated for the dataset. Based on the complex-

ity of the data and the need for capturing non-linear

patterns, Artiﬁcial Neural Networks (ANNs) were se-

lected as the most suitable model.

ˆy = f

∑

i=1

· x

+ b

(3)

where ˆy is the predicted output, w

are the weights,

are the inlets, and b is the bias-term.

3.6.2 Model Training and Validation

ANN model was trained on the processed dataset,

consisting of features such as CO, NH

, SO

, H

con-

centrations, temperature, and humidity. The model

was conﬁgured with a single hidden layer comprising

64 neurons and ReLU activation functions. The out-

come section employs the linear activation function

for regression.

Loss Function: MSE =

∑

i=1

− ˆy

)

(4)

is the true value, ˆy

is the anticipated outcome,

and n is the count of samples

3.6.3 Performance appraisal

The framework interpretation was reckoned using

RMSE (Root Mean Squared Erro and MAPE (Mean

Absolute Percentage Error). The effects were com-

pared against traditional models to demonstrate the

superior accuracy of the ANN model.

RMSE =

∑

i=1

− ˆy

)

(5)

MAPE =

∑

i=1



− ˆy



× 100 (6)

3.7 Comparative Analysis

To validate the suggested model, its performance was

likened to additional strategies such as Linear Regres-

sion, Polynomial Regression, and LSTM. The ANN

model demonstrated superior performance, with a

lower RMSE and MAPE, and higher prediction ac-

curacy.

4 RESULTS

The results presented here are derived from the im-

plementation and testing of the proposed air quality

monitoring and prediction system. Our approach, as

outlined in the Proposed Work section, integrates IoT

sensors, cloud-based data storage, and machine learn-

ing for effective air quality monitoring.

4.1 Sensor Data and AQI Analysis

The study performed a sequel of tests to validate the

sensor data and its integration with the ThingSpeak

Breath of the Future: Predicting Air Quality Index with ML and IoT

725

cloud. The data collected from the proposed mi-

crocontroller setup using the ThingSpeak API is dis-

played in real-time through the ThingSpeak control

panel. This includes continuous monitoring of key

pollutants: NH3, SO2, and CO.

Figure 3: Time-Series Analysis of NH3 Levels

The chart 3 shows NH3 levels over time, with a

sharp peak around 11:25, followed by a gradual de-

cline.

Figure 4: Trends in SO2 Concentrations Over Time

The diagram 4 depicts a sharp spike in sulfur diox-

ide (SO2) levels around 11:25 AM, followed by a

rapid decline. SO2 concentration remained relatively

low before and after this peak.

Figure 5: Temporal Variation of CO Levels

The graph 5 shows CO levels over time. The lev-

els remain relatively stable until 11:18, where they

drop sharply, then rise again to a peak at 11:20 before

falling back down.

The sensor data is then used to calculate AQI val-

ues, which provide a snapshot of air quality over a

given period. Table 1 shows the calibrated values of

all parameters, including temperature and humidity,

after proper calibration of the sensors. The calcu-

lated AQI values categorize air quality from ”Good”

to ”Hazardous” for the entire day.

Table 1: Data for NH

, SO

, and CO

23.48 24.55 0.97

24.00 25.50 0.98

8.00 6.20 0.12

12.85 4.87 1.28

13.80 5.65 0.58

4.2 Prediction Model Performance

To predict future pollutant levels, we employed Artiﬁ-

cial Neural Networks (ANNs), chosen for their capa-

bility to model complex data patterns. The prediction

model was trained and validated using the collected

dataset, leading to a signiﬁcant Root Mean Squared

Error (RMSE) of 82.84 and a high classiﬁcation suc-

cess rate of 94.54%. These metrics demonstrate the

model’s robustness and effectiveness in forecasting

air quality based on sensor data.

Table 2 shows the predicted AQI values gener-

ated by the ANN model. The results indicate that

the model accurately predicts AQI levels, aligning

closely with the actual sensor data, thereby validating

the model’s reliability.

Table 2: Predicted AQI Values

CO AQI

23.48 24.55 0.97 290.85663

24.00 25.50 0.98 295.90050

8.00 6.20 0.12 58.04753

12.85 4.87 1.28 210.23013

13.80 5.65 0.58 54.69106

The model’s architecture, developed using Ten-

sorFlow, incorporates layers with speciﬁc activation

functions and regularization techniques to optimize

performance. The accuracy graph (Fig. 6) illustrates

the model’s learning progression, conﬁrming its abil-

ity to classify AQI levels effectively.

These results not only substantiate the proposed

methodology but also emphasize the model’s capac-

ity to contribute signiﬁcantly to air pollution moni-

toring and forecasting. The high accuracy and low

INCOFT 2025 - International Conference on Futuristic Technology

726

Figure 6: Model Accuracy Over Training Epochs

error rates highlight the potential for deploying this

system in real-world applications, enhancing both en-

vironmental monitoring and public health awareness.

5 CONCLUSIONS

The study has made signiﬁcant progress in advancing

AQ Oversight and anticipation by leveraging the har-

monious combination of Machine Learning and the

Internet of Things. The consequences demonstrate

the potency of the designed system, particularly in uti-

lizing Artiﬁcial Neural Networks (ANNs) for predict-

ing Air Quality Index (AQI) with high accuracy. The

prototype achieved an RMSE of 82.84 and a classiﬁ-

cation precision of 94.54%, underscoring its capabil-

ity to capture complex patterns in air quality data.

The comprehensive system, which combines so-

phisticated hardware conﬁgurations with advanced

software algorithms, presents a dynamic and efﬁcient

approach to environmental monitoring. This inno-

vation enhances our comprehension of air pollution

dynamics and even enables preventive environmental

management strategies. The real-time data acquisi-

tion facilitated by IoT devices, coupled with the pre-

dictive analytics provided by ML, shows immense po-

tential in addressing the critical challenges of air pol-

lution.

As global industrialization and urbanization con-

tinue to intensify, the insights and methodologies de-

veloped in this study contribute meaningfully to the

ongoing global discourse on sustainable environmen-

tal practices. By harnessing the power of advanced

technologies, the points the path toward a destiny

where predictive modeling and real-time monitoring

work in concert to safeguard human health and pro-

tect ﬂimsy ecosystems. The ﬁndings highlight the

importance of continued innovation and shared com-

mitment to creating a healthier and cleaner planet for

future generations

REFERENCES

Rakib, M., Haq, S., Hossain, M. I., and Rahman,

T., 2022. IoT Based Air Pollution Monitoring

& Prediction System. In Proceedings of the 3rd

International Conference on Innovations in Sci-

ence, Engineering and Technology (ICISET),

Chittagong, Bangladesh, pp. XX-XX.

Mihirani, M., Yasakethu, L., and Balasooriya,

S., 2022. Machine Learning-based Air Pollu-

tion Prediction Model. Sri Lanka Technological

Campus, School of Engineering and Technology.

Gupta, N. S., Mohta, Y., Heda, K., Armaan, R., Valar-

mathi, B., and Arulkumaran, G., 2023. Predic-

tion of Air Quality Index Using Machine Learn-

ing Techniques: A Comparative Analysis. Jour-

nal of Environmental and Public Health, Volume

2023, Article ID 4916267, pp. XX-XX.

endez, M., Merayo, M. G., and N

nez, M., 2023.

Machine Learning Algorithms to Forecast Air

Quality: A Survey. Published online: 16 Febru-

ary 2023.

Bhattacharya, S., and Shahnawaz, S., 2022. Using

Machine Learning to Predict Air Quality Index

in New Delhi. Jadavpur University, Department

of Computer Science and Engineering.

Gogineni, A. C., and Murukonda, V. S. N. M., 2022.

Prediction of Air Quality Index Using Super-

vised Machine Learning.

Kulkarni, M., Rajule, N., Raut, A., and Pawar, S.,

2022. Air Quality Monitoring and Prediction us-

ing SVM. Dr. D. Y. Patil Institute of Technol-

ogy, Department of Electronics & Telecommuni-

cation Engineering, Pimpri, Pune, India.

Sonawane, P., Dhanawade, S., Barangule, V., Kulka-

rni, A., and Mahalle, P., 2022. Air Quality

Analysis & Prediction Using Machine Learning:

Pune Smart City Case Study. Vishwakarma In-

stitute of Information Technology, Dept. of Me-

chanical Engineering, Pune, India.

Murugan, R., and Palanichamy, N., 2021. Smart City

Air Quality Prediction using Machine Learning.

In Proceedings of the 5th International Confer-

ence on Intelligent Computing and Control Sys-

tems (ICICCS 2021), IEEE Xplore, ISBN: 978-

0-7381-1327-2.

Kasetty, S. B., and Nagini, S., 2022. A Survey Pa-

per on an IoT-based Machine Learning Model

to Predict Air Pollution Levels. In Proceedings

of the 4th International Conference on Advances

in Computing, Communication Control and Net-

working (ICAC3N 2022).

Breath of the Future: Predicting Air Quality Index with ML and IoT

727

Krishna, Y. S., and Nabi, M. A., 2022. Prediction of

Air Pollutants Using Supervised Machine Learn-

ing. Bachelor’s Thesis, Department of Computer

Science and Engineering.

Zhang, D., and Woo, S. S., 2020. Real-Time Lo-

calized Air Quality Monitoring and Predic-

tion Through Mobile and Fixed IoT Sens-

ing Network. IEEE Access, DOI: 10.1109/AC-

CESS.2020.2993547.

Kumari, A., Vasuki, H. R., Kumar, K. S. A., Nikesh,

M. P., and Raju, H. V., 2020. Prediction of Air

Quality in Industrial Area. In Proceedings of the

5th International Conference on Recent Trends

on Electronics, Information, Communication &

Technology (RTEICT-2020), Bangalore, India,

November 12-13.

Gore, R. W., and Deshpande, D. S., 2022. An Ap-

proach for Classiﬁcation of Health Risks Based

on Air Quality Levels. Marathwada Institute of

Technology, Department of Computer Science

and Engineering, Aurangabad, Maharashtra.

Shaban, K. B., Kadri, A., and Rezk, E., 2022. Urban

Air Pollution Monitoring System With Forecast-

ing Models.

INCOFT 2025 - International Conference on Futuristic Technology

728