XAI-Enhanced Forecasting: Deep CNN- LSTM for Air Quality

Accurate air quality forecasting is vital for public health and urban policy-making. Conventional statistical approaches, including ARIMA fail to capture nonlinear patterns but algorithms like Random Forest and XGBoost, despite their strengths, fail to represent long-term sequential relationships in data.. Deep learning approaches, including LSTM and CNN hybrids, improve accuracy but are limited by poor interpretability. This work proposes a hybrid DCNN- LSTM framework integrated with SHAP for explainability. Experiments on Delhi air quality data demonstrate that the model achieves superior accuracy (R² = 1.000, RMSE = 5.74) while identifying PM2.5, NO₂, and SO₂ as key pollutants, thereby ensuring both predictive performance and transparency.