Forecasting ENSO Impact on India’s Economic Indicators Using AI and Climate Data: A Cross-Model Evaluation

This research explores whether adding climate signals improves short- horizon predictions of India’s macroeconomic variables, & determines which algorithm generates the most tremendous gains in accuracy. Using monthly data for January 2010–December 2024, it examines six targets—GDP growth, CPI inflation, IIP growth, unemployment, NIFTY-50 returns, and an agricultural input/output ratio—augmented with El Niño–Southern Oscillation (ENSO) indices and all-India rainfall anomalies. A cross-model framework compares Long Short-Term Memory networks (LSTM), Extreme Gradient Boosting (XGBoost), and Least Absolute Shrinkage and Selection Operator (LASSO) under identical features, splits, and horizons (nowcast, +1, +3 months), evaluating RMSE, MAE, and directional accuracy. Explainability uses permutation importance and SHAP; ablation isolates the marginal value of rainfall and ENSO; regime tests quantify asymmetry across dry/normal/wet months. Results show climate augmentation improves accuracy most for CPI and agriculture (ΔRMSE ≈ 9– 13% and 13–15% at +1 month), followed by GDP (≈ 6–11%) and smaller but non- trivial gains for IIP and unemployment. LSTM yields the lowest errors for CPI and GDP, while XGBoost performs best for IIP and unemployment; LSTM modestly outperforms for NIFTY. Improvements are state-dependent and largest in dry months (e.g., CPI ≈ +18.7% ΔRMSE), with rainfall contributing more than ENSO, though both are additive. The findings support indicator-specific model choice and regime-aware monitoring. Policy applications include improved inflation nowcasting, anticipatory food-management operations, and targeted labour-market support during rainfall deficits