Residual Stacking with Temporal Convolutional Network and Light Gradient Boosting for Air Quality Index Prediction

Reliable forecasting of the Air Quality Index (AQI) is vital for early public-health action and urban management.This study introduces a two-stage ensemble in which a Temporal Convolutional Network (TCN) captures short-term pollutant trends and a Light Gradient Boosting Machine (LightGBM) corrects its residuals using contextual cues such as festival periods, monsoon onset, and pollutant deviations.On Indian hourly data (2021-2024), the system achieved mean absolute and root-square errors of 19.4 and 28.5 for six-hour forecasts-an 8-15 % improvement over strong baselines.During the Diwali episode, the model reproduced extreme AQI peaks with 81 % recall.The results indicate that residual stacking strengthens predictive stability under irregular conditions while remaining computationally light and easily interpretable.