Explaining INR Exchange Rate Forecasts with XGBoost and SHAP: An Interpretable Machine Learning Framework

Accurate exchange rate forecasting is essential for trade, investment, and policy planning. This study uses an XGBoost model to predict INR exchange rates against USD, EUR, GBP, and JPY, based on monthly data from May 1966 to February 2025. The model showed strong performance with RMSE/MAE scores of 15.35/13.58 (USD), 6.36/5.62 (EUR), 12.42/10.87 (GBP), and 0.03/0.02 (JPY). Using SHAP values, the model's predictions were explained clearly, highlighting that 1-month and 3-month lagged rates had the most impact. This aligns with financial market behavior and adds trust to the model. Visual comparisons between predicted and actual rates showed the model could capture historical trends well, making it a practical tool for analysts, policymakers, and businesses. Future improvements could include economic indicators or deep learning methods like LSTM for long-term forecasting.