Enhanced air quality index classification: leveraging genetic algorithm and SMOTE for accurate assessments in Indian cities

Abstract

The escalating air pollution levels in Indian metropolitan regions necessitate robust predictive systems for air quality assessment. This study presents an advanced air quality index (AQI) forecasting model leveraging the radial basis function (RBF) kernel-based extreme learning machine (ELM) optimized using a genetic algorithm (GA). The proposed model is evaluated on real-time AQI datasets from four major Indian cities: Vishakhapatnam, Delhi, Hyderabad, and Patna. Initial experiments without class balancing yielded prediction accuracies of 83.9%, 88.3%, 87.0%, and 86.9% respectively. To address the class imbalance and enhance predictive performance, the synthetic minority oversampling technique (SMOTE) was applied. Post-balancing, the model achieved significantly improved accuracies of 93.9%, 94.7%, 92.3%, and 96.2% across the respective cities. These results underscore the effectiveness of integrating SMOTE with RBF-ELM for AQI prediction and demonstrate the critical role of data balancing in improving model generalizability. The proposed approach offers a promising solution for urban air quality monitoring and can assist policymakers in formulating timely interventions to mitigate health risks associated with air pollution.