Cotton Mitra: An Explainable Ensemble Forecasting and Supply-Chain Intelligence Platform for India's Cotton Sector

Abstract

Supply chain efficiency in India's textile and agricultural sectors mainly relies on correctly predicting the demand for cotton. Cotton Mitra is a tool that helps make better decisions by using artificial intelligence. It combines several features like predicting product demand, managing inventory with ICT software, providing market insights through MIPS, and offering advice to farmers. Four tree-based machine learning models are used together in the forecasting system—XGBoost , Random Forest, Extra Trees, and Gradient Boosting—by applying optimized weights for combining them, which are found using Sequential Least Squares Programming (SLSQP). It has a module for error correction that uses ARIMA to pick up on any remaining time-based patterns and makes the prediction more accurate. This study uses 184 monthly data points gathered from official government and commodity market sources from January 2011 up to April 2026. This helps keep the timing consistent and stops information from leaking into the wrong part of the data by using Walk-forward TimeSeriesSplit for validation. SHAP-based explanations help explain how a specific model works by providing detailed insights into its decision-making process. Experiments show good performance under experimental conditions - R2 0.9412, MAPE 3.87%, RMSE 0.38 lakh bales.? By integrating with the platform, users can monitor their inventory and access market analytics or assist farmers through AI-powered assistance, offering an effective and interpretable solution for data supply chain decision-making in the Indian cotton sector.