Maximum Overlap Discrete Wavelet Packet Transform Based Deep Convolutional Neural Networks for Predicting Ambient Air Pollutants’ Concentration

This research proposes an optimized Deep Convolutional Neural Network to predict PM10, PM2.5, NO2, and O3 levels across 32 locations in Delhi. The model integrates a hybrid Grasshopper and Seagull Optimization Algorithm to enhance predictive accuracy. To manage complex time-series data, the Maximum Overlap Discrete Wavelet Packet Transform was employed for multi-scale decomposition. Using data from 2009 to 2010, the model achieved high-precision results, notably for PM2.5 with an RMSE of 0.087 and an Index of Agreement of 0.999. These findings demonstrate the model’s effectiveness in forecasting high-pollutant events to mitigate community health risks.