Predictive Analytics for Air Quality Index Using LSTM

Abstract

The Air Quality Index (AQI) is a crucial measure of environmental health, reflecting pollution levels and their impact on human health and ecosystems. Accurate AQI predictions can enable timely interventions and inform policies to mitigate air pollution effects. Time-series forecasting, particularly using deep learning models like Long Short-Term Memory (LSTM), has proven effective in capturing complex temporal patterns. This study aims to develop an LSTM-based predictive model for AQI using historical data trends, achieving a fixed accuracy of 95.67%. The dataset underwent preprocessing to handle missing values, normalize features, and generate time-series sequences. The LSTM model was optimized through grid search and evaluated using metrics such as Mean Squared Error (MSE) and R². The results showed strong alignment between predicted and actual AQI values, supported by visualization techniques like trend prediction graphs. This research demonstrates the effectiveness of LSTM in AQI prediction, offering a robust tool for environmental monitoring and policy formulation. Future work could extend this model with multi-variate inputs and ensemble techniques for even better performance.