A Novel Approach for Air Quality Index Prognostication using Hybrid Optimization Techniques

This research presents an innovative deep learning approach for forecasting the Air Quality Index (AQI), a crucial public health concern in both developed and developing countries. The proposed methodology encompasses four stages: (a) Pre-processing, involving data cleaning and transformation; (b) Feature Extraction, capturing central tendency, dispersion, higher order statistics, and Spearman's rank correlation; (c) Feature Selection, using a novel hybrid optimization model, Particle Updated Grey Wolf Optimizer (PUGWO); and (d) an ensembled deep learning model for AQI prediction, integrating a Convolutional Neural Network (CNN), an optimized Bi-directional Long Short-Term Memory (Bi-LSTM), and an Auto-encoder. The CNN and Auto-encoder are trained on the extracted features, and their outputs are fed into the optimized Bi-LSTM for final AQI prediction. Implemented on the PYTHON platform, this model is evaluated through R^2, MAE, and RMSE error metrics. The proposed HRFKNN model demonstrates superior performance with an R-Square of 0.961, RMSE of 11.92, and MAE of 10.29, outperforming traditional models like Logistic Regression, HRFLM, and HRFDT. This underscores its effectiveness in delivering precise and reliable AQI predictions.