Comparative Analysis of Air Quality Index Prediction Using Machine Learning Techniques

As technology and development accelerate at a rapid rate, pollution is a growing concern. In this study, a comparative analysis of two neural network models for the prediction of Air Quality Index (AQI) was performed. Pollutant concentration data were collected from 26 Indian cities to explore the effectiveness of Multi-Layer Perceptron (MLP) and Radial Basis Function Neural Network (RBFNN) models in predicting AQI. The MLP model, comprising two hidden layers with ReLU activation and trained using the Adam optimiser, achieved a testing $\mathbf{R}^{\mathbf{2}}$ score of $\mathbf{0. 8 8 9 3}$, a Mean Absolute Error (MAE) of 0.1603, a Mean Squared Error (MSE) of 0.1057, and a time of 8.3 seconds. The RBFNN model utilised a single hidden layer with Gaussian activation, with its centers determined using the method of K-means clustering and the Elbow method. This model attained a testing $\mathbf{R}^{\mathbf{2}}$ score of $\mathbf{0. 8 7 6}$, RMSE of $\mathbf{0. 3 4 6 4}$, MSE of 0.12, and a time of 11 seconds. Based on the results, while MLP provided superior performance on average, RBFNN offered consistent predictions across a wider AQI range. By integrating machine learning models with IoT devices, there is scope for significant improvement in real-time AQI prediction and forecasting.