AI‑Driven IoT Framework for Real‑Time Air Quality Monitoring

and Stress Correlation Analysis

R. Manikandan, Gobinath G., Gokul G. and Boobalan D.

Department of Computer Science and Engineering, Nandha Engineering College (Autonomous), Erode, Tamil Nadu, India

Keywords: PM2.5, CO2, IOT, Stress, AI‑driven.

Abstract: Burden, air pollution, and stress in homes, offices, and urban settings. This project strikes at the nexus between

environmental monitoring and mental health, given that air pollution is a significant global challenge, it has

far-reaching implications on physical and mental health. The negative effects of poor air quality on the

respiratory and cardiovascular systems are by now well established, but the same cannot be said for the impact

of its phytotoxicity on the psyche, especially in real-time, localized contexts. Here, we put forward an AI-

driven IoT framework for real-time air quality monitoring and stress prediction, which provides users with

actionable intelligence without requiring any manual input. The system is a combination of low-cost IoT

sensors (PM2. 5, CO2) using an ESP32 microcontroller to acquire data from the environment in real time and

transmit it to a backend hosted in the cloud for analysis. It uses a machine learning model trained on a dataset

with information linking air pollution and metrics of stress to predict stress, based on current pollution levels.

The anticipated stress levels, along with air quality data, are shown on a mobile app, which also provides

recommendations tailored to the user (for instance, “Open windows,” “Avoid outdoor activities”) and issues

alerts when air quality worsens or stress levels are expected to increase. Utilizing the advances in artificial

intelligence according to Internet of Things this system provides a scalable solution for monitoring the both

contributing towards smart health technologies and promote living in sustainable way.

1 INTRODUCTION

Air pollution is one of the biggest global challenges

we are facing nowadays and its adverse effects on

physical health have been well-studied 1(Neeraja et.,

al.2024). However, its psychological effects,

including increased stress, stress and cognitive decline

are largely ignored especially at real time context

studies of localized level 9. Air quality monitoring

with a focus on environmental parameters is not a new

concept, although its study from the point of view of

mental well- being is relatively unexplored 3. To

mitigate this gap, this project proposes an AI- driven

IoT framework that enables the continuous monitoring

of the air quality in real-time and cross-relates it with

the stress levels, providing insights without taking up

manual user efforts.

The proposed system uses inexpensive IoT sensors

(e.g., PM2. 3, (S. Subha et., al. 2024)) equipment and

ESP32 microcontroller to record your hyperlocal air

quality data 1[ 6. javascript:;]. This information is then

sent out to a cloud-based system which uses a machine

learning model that was previously fed historical

datasets connecting pollution levels and psychological

stress to predict user well-being 2. The system is built

with seamless usability by automating stress

detection, providing alerts and recommendations

through a mobile interface in real-time. For example,

declining air quality will send alerts such as “High

PM2. 5 detected air out,” while predictive stress

analytics encourage proactive measures such as “Take

a break stress likely to peak in 30 minutes” 7.’

Three Hurdles in Solutions Addressed by this

Innovation: Proactive Health Monitoring: An AI 2.

Scalable deployment 1: Uses easy to purchase

commercial hardware. Disciplinary Infraction:

Leverages environmental science with mental health

analytics for holistic well-being 3.

This framework not only enables seamless

integration of

IoT, AI, and user-centric design in

smart health technologies, but also sets a new

benchmark in various application domains, including

urban planning, workplace wellness, and community

health programs 4[2]. IoT Sensor and Cloud-Based

Analytics Fusion: for Hyper-Local Air Quality

Monitoring

1(Neeraja et., al.2024).

Manikandan, R., G., G., G., G. and D., B.

AI-Driven IoT Framework for Real-Time Air Quality Monitoring and Stress Correlation Analysis.

DOI: 10.5220/0013892200004919

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

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 3, pages

87-92

ISBN: 978-989-758-777-1

Stress Correlation Analysis: Predicts level of stress

based on pollution data using machine learning 2.

Actionable Insights: Provides tailored suggestions

and alerts through a mobile application 7.

2 RELATED WORKS

2.1 Real-Time Air Quality Monitoring:

A Smart IoT System Using Low-

Cost Sensors and 3D Printing

This work has designed a portable air quality station,

contained in a 3D-printed case, with the aim of

simplifying data collection and reducing material

usage in an experimental laboratory. As for air

contaminants and gases of concern, with implications

for susceptible populations (e.g., asthmatics and

children), this breakthrough holds great promise for

public health. Regarding the role of indoor

ventilation, as pointed out by the COVID-19

pandemic, which causes the infection through

airborne particles, the importance of effective

surveillance and preventive measures cannot be

overstated. The station is based on open-source

Python software, a Raspberry Pi core data

collection/storage platform interfacing through

GPIO, serial, and I2C interfaces with the sensors. The

device is structurally modular so that measurements

and target pollutant can be modified by the user.

Validation with end-user testing established the

effectiveness and usability of the system in real world

applications. The mobile infrastructure provides a

low-cost alternative for air quality network

development to meet the needs of

disadvantaged/vulnerable communities. The module

exhibited a high reliability of 95.30% in identifying

ubiquitous pollutants, confirmed by CO2 level

assessment in classrooms (90.47% reliability (Osa-

Sanchez and B. Garcia-Zapirain, 2025). in

comparison to commercially available devices) and

by air quality assessment in the air in the environment

(85.63% reliability (Osa-Sanchez and B. Garcia-

Zapirain, 2025).

2.2 Mental Fitness Tracker using

Regression Models

Mental health is of considerable importance in human

health, and it has implications across many areas of

life. Yet, assessment and enhancement of mental

well-being is not a trivial matter as there are

numerous complexes, heterogeneous aspects. In this

work, we use artificial intelligence (AI) technology,

as a means to suggest a new mental health cognitive

tracking and enhancement. As a solution, mental

fitness tracker (MF tracker) is a web-based tool that

monitors and analyzes a symptomatic user's

behaviors, emotions, and mental state providing

information to the user from social media, wearable

devices, electronic surveys, and natural language

processing. Based on the data analysis, the program

provides users with personalized suggestions, advice

and action items in order to improve their mental

health. (P. Malin Bruntha et., al. 2024).

3 METHODOLOGY

The suggested system is constructed with the ability

to track real-time air quality as well as stress level by

leveraging an AI-enabled IoT environment. The

methodology is subdivided into 5 major stages in

order to guarantee an harmonized integration of

hardware, software and analytics. Figure 1 shows the

workflow.

Figure 1: Workflow.

3.1 Data Collection and Preprocessing

3.1.1 Predefined Dataset

Air Quality Data: The historical air quality data such

as PM2. 5 and CO₂ levels, which were extracted

from U.S. EPA AirData and OpenAQ API

(Osa-

Sanchez and B. Garcia-Zapirain

, 2025). 4. The

dataset consists of hourly data from different

locations, grouped

into air quality categories (AQI)

ranging from Good, Moderate, and Unhealthy.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

Stress Based Data: Based on Air Quality metrics

Fake data set was generated to simulate modifies

stress level 1-5. For instance, PM2. Levels higher

than 50 µg/m³ were correlated with stress levels

higher than 3, in accordance with the well-established

relationship between pollution and mental health

well-being 2.

3.1.2 Data Preprocessing

Normalization: PM2. The 5 and CO₂ values were

standardized with Min-Max scaling to be consistent

across the dataset 5.

Handling of outliers: Interpolating techniques

were used to fill missing values while trying to

maintain the integrity of the data (J. Pellegrino et., al.

2025).

Feature Engineering: Additional features were

created based on the time of day (average measured

values for each hour, peak pollution valleys– e.g.:

morning peaks, day highs, etc.) to increase models’

accuracy 3.

3.2 AI Model Development

3.2.1 Stress Prediction Model

Data, Models and Procedures: In order to identify the

best algorithms, we used a Random Forest Regressor,

as they can handle non-linear relationships as well as

assess feature importance 2.

Training: 80% of the data was used for training

and the other 20% was kept for testing.

Hyperparameters such as n_estimators and

max_depth was optimized using Grid Search 2.

Evaluation: The R² score for the model was 0.78,

demonstrating a high accuracy level for predicting

stress levels from the air quality data 2.

3.2.2 Air Quality Classification

Algorithm: A Long Short-Term Memory (LSTM)

network was used for time-series forecasting of

PM2.5 trends (J. Rosa-Bilbao et., Al. 2025). (Neeraja

et., al.2024). Architecture: The model included two

LSTM layers (64 units each) with a dropout rate of

0.3 to prevent overfitting (J. Rosa-Bilbao et., Al.

2025). (Neeraja et., al.2024).

Training: The model was trained on 7 days of

historical data and validated on the next 24 hours,

achieving a Mean Absolute Error (MAE) of 0.45 (J.

Rosa-Bilbao et., Al. 2025). (Neeraja et., al.2024).

3.3 IoT System Integration

3.3.1 Hardware Setup

Sensors: MQ135 (CO₂), PMS5003 (PM2.5), and

DHT22 (temperature/humidity) were connected to an

ESP32 microcontroller (Osa-Sanchez and B. Garcia-

Zapirain, 2025). (J. Pellegrino et., al. 2025) (Neeraja

et., al.2024). Data Transmission: The ESP32’s Wi-Fi

module sent JSON payloads to the Firebase Realtime

Database every 60 seconds (J. Pellegrino et., al.

2025). (Neeraja et., al.2024).

3.3.2 Firmware Development

The firmware, developed in the Arduino IDE,

included error- handling mechanisms such as

recalibration loops for the MQ135 sensor (Osa-

Sanchez and B. Garcia-Zapirain, 2025). (J. Pellegrino

et., al. 2025). A watchdog timer was implemented to

ensure system stability during prolonged operation (J.

Pellegrino et., al. 2025). (Neeraja et., al.2024).

3.4 Mobile App Development

3.4.1 Platform

The mobile app was built using Thunkable, a no-code

platform, and integrated with Firebase for real-time

data synchronization (Aubakirov et., al. 2024) (K.

Ramar et., al. 2022).

3.4.2 Features

Real-Time Dashboard: Displays PM2.5, CO₂, and

predicted stress levels using intuitive gauges and line

charts (Aubakirov et., al. 2024) (K. Ramar et., al.

2022).

Notifications: Firebase Cloud Messaging (FCM)

sends alerts when PM2.5 exceeds WHO thresholds

(25 µg/m³) or stress levels are predicted to rise

(Aubakirov et., al. 2024) (K. Ramar et., al. 2022).

Recommendations: Rule-based logic provides

actionable feedback, such as “PM2.5 > 30 → ‘Close

windows’” (Aubakirov et., al. 2024) (K. Ramar et., al.

2022).

3.5 System Evaluation

3.5.1 Performance Metrics

AI Model: Achieved a MAE of 0.45 for stress

prediction (

P. Malin Bruntha et., al. 2024). (Aubakirov

et., al. 2024) . IoT System: Demonstrated a latency of

less than 2 seconds for data transmission

(J. Pellegrino

AI-Driven IoT Framework for Real-Time Air Quality Monitoring and Stress Correlation Analysis

et., al. 2025). (Neeraja et., al.2024). User Feedback: A

pilot test with 20 users achieved 90% accuracy in

actionable recommendations (Aubakirov et., al.

2024) (K. Ramar et., al. 2022).

3.5.2 Statistical Validation

Pearson Correlation: A strong correlation (r = 0.65, p

< 0.01) was found between PM2.5 levels and stress

levels (P. Malin Bruntha et., al. 2024). (Aubakirov et.,

al. 2024).

ANOVA Test: Significant differences in stress

levels across air quality categories were confirmed (F

= 12.7, p < 0.001) (P. Malin Bruntha et., al. 2024).

(Aubakirov et., al. 2024).

3.6 Key Contributions

Real-Time Monitoring: Combines IoT sensors and

cloud- based analytics for hyper-local air quality

tracking [1] (J. Pellegrino et., al. 2025). (Neeraja et.,

al.2024).

Stress Correlation Analysis: Uses machine

learning to predict stress levels based on pollution

data (P. Malin Bruntha et., al. 2024). (Aubakirov et.,

al. 2024).

Actionable Insights: Delivers personalized

recommendations and notifications via a mobile app

(Aubakirov et., al. 2024). (K. Ramar et., al. 2022).

Table 1 represents the air quality index.

Table 1: Air Quality Index.

AIR QUALITY

INDEX

CATEGORY

0-50 Good

51-100 Satisfactory

101-200 Moderate

201-300 Poor

301-400 Very Poor

401-500 Severe

3.7 Random Forest Accuracy

Figure 2: Accuracy of Random Forest.

Figure 2 shows the Performance analysis of

regression models through R² score (coefficient of

determination) If signifies the response variable

(e.g., levels of stress) and the independent variables

(e.g., PM2. 5, CO2 levels).

The R² score of 0.78 shows that 78% of the

changing stress level can be accounted for by

changing air quality metrics (PM2. 5 and CO2). The

model's ability to explain high percentage of balance

reflects a successful representation of the relationship

between these environmental variables and

components of psychological wellbeing. (78% of

variation explained) and the remaining 22% could be

attributed to the othermodel components.

The closer the R² value is to 1, the better the

predictive power and vice-versa. The appropriateness

of this model for predicting associations between air

quality and stress is reflected in the R² score of 0.78,

showing that this AI-model is applicable toward the

monitoring and -real-time intervention of stress

(Fazel, Liu et al. 2020).

3.8 LSTM Model Accuracy

Figure 3: Accuracy of LSTM Model.

Figure 3 shows the mean absolute error is mean

absolute error or mae is a common loss function for

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

regression models and by assuming the assumption

that 50 gm is the true pm2 5 level this error indicates

the error for each blazer or block 5 pm2 5 reading a

lower mae indicates more accurate models because

the error of predicting.

The predicted pm2 5 mean error of the models to

the true pm2 5 reading equals 045-unit 50 gm

specifically in lst air quality large systems to monitor

air quality the maen is 045 during pm2 forecasting

during lst 50 gm was defined as the true pm2 5 level

with 045 degree meaning this was a significant degree

of power in the simulation architecture trend

senabling instantaneous interventions.

Such as alerts or recommendations when

pollution levels spike while all of the above methods

are significant for robust basic readings the

disadvantage is that neither are quite-high enough and

so not perfect for real-time use where the answer must

be checked for immediate use and hence they all need

a confirmation or a different precision to prospect the

state of the human.

4 EXPERIMANTAL RESULT

Figure 4: Predicted Vs Actual Stress Level.

Figure 4 shows the predicted vs actual stress level.

X-axis: Actual Stress Levels (1–5 scale, collected

from synthetic/user data).

Y-axis: Predicted Stress Levels (1–5 scale, output

by the Random Forest model).

Trend Line: A diagonal line (y = x) represents

perfect predictions.

Data points cluster closely around the trend line

indicating strong alignment between predicted and

actual valuesr 078 confirms that 78 of stress

variability is explained by air quality metrics pm25 co

Data on air quality only the measured data from april

114 2016 are selected for further analysis for example

the air quality data which includes pm25 pm10 and

aqi are shown in figure 5 to confirm the accuracy of

this air quality monitoring system another set of data

is taken from a well-known pm25 historical database

namely the young-0com 13data on air quality only

the measured data from april 114 2016 are selected

for further analysis for example the air quality data

which includes pm25 pm10 and aqi are shown in fig

5 to confirm the accuracy of this air quality

monitoring system another set of data is taken from a

well-known pm25 historical database namely the

young-0com 13, i.e., the Young-0.com .. (K. Zheng

et., al. 2016).

Figure 5: Air Quality Data.

5 FUTURE ENHANCEMENTS

5.1 Multimodal Mental Health

Correlation

It combines the data with wearables EG heart rate

monitors EEG headbands to measure.

Physiological indicators of stress. eg heart rate

variability brainwave patterns and air quality data for

more holistic evidence of mental health impacts.

Expand dataset for socioeconomic factors, eg

income levels occupation and behavioral data, eg

sleep patterns physical activity to help refine

predictions of stress personalized.

5.2 Personalized Recommendations

Recommendations use reinforcement learning based

on user habits to personalize user feedback eg, jog in

low population routes gamify.eg, offer reward points

for reduced carbon footprints etc. sustainable

behavior.

6 CONCLUSIONS

By integrating various ai and IOT technologies this

AI-Driven IoT Framework for Real-Time Air Quality Monitoring and Stress Correlation Analysis

project demonstrates how to monitor your

environment and improve your mental health in real-

time. It helps evaluate air quality data and correlate it

with moments of stress or anxiety using inexpensive

sensors an accurate stress prediction model has been

created and machine learning has significantly

improved with machine learning r 078 mae 045

suggesting that it is feasible as a scalable user-

friendly tool for proactive health management by

offering prompt preventative advice. Such as

ventilation alert ambient stress prevention

recommendations it enables individuals and

communities to address the twin problems of air

pollution and mental stress. Supporting new behavior

this cross-sector integration can thus cement IOT ai

innovation but simultaneously find solutions to some

of the concerns raised by the un sustainable

development goals multimodal data crafting AI

model complexity and global expansion in future

works can increase the coverage of the system and

allow it to be a necessary help in smart city

infrastructure the study spotlights the importance of

multifaceted solutions to modern-day health

challenges especially SDG 3 good health and well-

being and SDG 11 sustainable cities by presenting on

both mental and environmental health on equal

footing.

REFERENCES

A. Osa-Sanchez and B. Garcia-Zapirain, "Real-Time Air

Quality Monitoring: A Smart IoT System Using Low-

Cost Sensors and 3D Printing," in IEEE Journal of

Radio Frequency Identification, doi: 10.1109/JRFID.

2025.35418

J. S. A R, P. Malin Bruntha, D. S, R. G, K. A and B. C.

Mary M, "Mental Fitness Tracker Using Regression

Models," 2024 3rd International Conference for

Innovation in Technology (INOCON), Bangalore,

India, 2024, pp. 1-5,

doi: 10.1109/INOCON60754.2024.10512049.

J. Rosa-Bilbao, F. S. Butt, D. Merkl, M. F. Wagner, J.

Schäfer and J. Boubeta-Puig, "IoT-Based Indoor Air

Quality Management System for Intelligent Education

Environments," in IEEE Internet of Things Journal,

doi: 10.1109/JIOT.2025.3539886.

J. Pellegrino, H. Aziza, M. Guerin, P. Taranto, W.

Rahajandraibe and B. Ravelo, "Development of a

Multi- Sensor Mobile Device for Urban Air Quality

Monitoring at the Street Corner: The SMILE Project,"

in IEEE Access, vol. 13, pp. 14857-14871, 2025

doi:10.1109/ACCESS.2025. 3530145.

R. Mckenzie et al., "A Cost-Effective Air Quality

Monitoring System for the Global South," 2024 IEEE

International Humanitarian Technologies Conference

(IHTC), Bari, Italy, 2024, pp. 1-7,

doi: 10.1109/IHTC61819.2024.10855074.

A. Neeraja, C. M. Begum, K. V. Reddy, T. V. Smitha and

N. Neelima, "Revolutionizing Air Quality Forecasting

in Bengaluru with Advanced Machine Learning

Techniques," 2024 First International Conference on

Innovations in Communications, Electrical and

Computer Engineering (ICICEC), Davangere, India,

2024, pp. 16,

doi:10.1109/ICICEC62498.2024. 10808771.

A. Aubakirov, K. Makhanov, A. Issainova, N. Glazyrina,

B. Razkhova and D. Satybaldina, "An IoT-Based Air

Pollution Monitoring System for Smart City," 2024 4th

International Conference on Computer Systems

(ICCS), Hangzhou, China, 2024, pp. 156-161,

doi:10.1109/ICCS62594.2024.10795835

S. Subha, N.Mohankumar, A. S. Rao, A. Ayub Khan, I. M.

V and S. Murugan, "IoT's Role in Personalized Dietary

Support for Mental Health using RNN Model," 2024

Third International Conference on Electrical,

Electronics, Information and Communication

Technologies (ICEEICT), Trichirappalli, India, 2024,

pp. 1-6, doi: 10.1109/ICEEICT61591.2024.10718428.

A. Aubakirov, K. Makhanov, A. Issainova, N. Glazyrina,

B. Razkhova and D. Satybaldina, "An IoT- Based Air

Pollution Monitoring System for Smart City," 2024 4th

International Conference on Computer Systems

(ICCS), Hangzhou, China, 2024, pp. 156-161, doi:

10.1109/ICCS62594.2024.10795835

S. A, S. R. LB, S. K. E, K. Ramar, H. S and B. A,"Mental

Health Tracker Using Machine Learning," 2022

International Conference on Disruptive Technologies

for Multi-Disciplinary Research and Applications

(CENTCON), Bengaluru, India, 2022, pp. 11-14, doi:

10.1109/CENTCON56610.2022.10051248.

R. S. S, V. S, P. R and A. Shiny, "Stress Relief and Anxiety

Therapy Using AI," 2024 International Conference on

Power, Energy, Control and Transmission Systems

(ICPECTS), Chennai, India,2024, pp.1-6,

doi:10.1109/ICPECTS62210. 2024. 10780170.

M. Venkatesan, S. Pawar and P. Thorat, "Depression Level

Detection Using AI,"

2024 International Conference on

Intelligent Systems and Advanced Applications

(ICISAA), Pune, India, 2024, pp. 1-5, doi:

10.1109/ICISAA62385.2024.10828795.

K. Zheng, S. Zhao, Z. Yang, X. Xiong and W. Xiang,

"Design and Implementation of LPWA-Based Air

Quality Monitoring System," in IEEE Access, vol. 4,

pp.32383245,2016,

doi:10.1109/ACCESS.2016.2582153.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES