Real-Time IoT-Based Toxic Gas Monitoring and Comparative Analysis of Machine Learning Techniques for Air Quality Index Prediction in Dhaka

Toxic gases pose health risks, including respiratory disorders. Traffic control and the identification of important pollutants are aided by the ability to predict the air pollution index. In order to predict the air quality index in Dhaka, Bangladesh, this study evaluates many machine learning techniques, including Support Vector Machine (SVM), KNearest Neighbors (KNN), Decision Tree (DT), Discriminant Analysis (DA), and Logistic Regression (LR). Furthermore, the study employs a sensor-based system to identify sulfur dioxide (SO2), methane (CH4), and ammonia (NH3) utilizing an Arduino Uno, NodeMCU ESP8266, MQ-4, MQ-135, and MQ-136 sensors. Experimental findings reveal significantly higher sensor readings in polluted air conditions: MQ-4 sensor (clean air: 0.0854 ppm, contaminated air: $\mathrm{1 2 0. 0 4 2 3} \mathrm{~ p p m}$), MQ-135 sensor (clean air: 0.0585 ppm, contaminated air: 140.184 ppm), and MQ-136 sensor (clean air: 0.0641 ppm, contaminated air: $\mathrm{1 0 0. 8 1 2 8 ~ p p m}$). Various data pre-processing techniques were employed prior to modeling using Kaggle’s Dhaka dataset. Among the models, Decision Tree (DT) yielded the highest performance metrics ($\mathrm{R}^{2}$ $=0.9999$, $\mathrm{MSE}=0.4410$, RMSE $=0.6641$), demonstrating superior predictive capabilities for Dhaka’s air quality index compared to other methods.