Prediction of Metal Prices Using Learning Machine Methods: Case of Indian Economy

Abstract

This study evaluates machine learning and deep learning methodologies for forecasting metal prices (Titanium PPI, Bauxite & Aluminum, Coal Australia, and Iron Ore) within the context of the Indian economy. To enhance predictive precision, the analysis integrates key macroeconomic indicators, including the Indian CPI, industrial production, and the U.S. Federal Funds Rate. Empirical results demonstrate that tree-based ensembles, specifically the Random Forest model, consistently outperform other tested algorithms by achieving the lowest Root Mean Squared Error (RMSE) across all four commodities. Despite the theoretical advantages of deep learning for temporal sequence modeling, our findings indicate that tree-based models provide superior generalization and robustness against market volatility in this specific context. Furthermore, SHAP value analysis reveals that autoregressive target lags and specific macroeconomic variables are the primary drivers of price forecasts. These insights offer actionable guidance for policymakers and industrial stakeholders engaged in strategic planning and risk management.