Agricultural decision-making in the Bargarh Canal command (BCC) of eastern India faces challenges due to climate variability and resource limitations. This paper presents an advanced machine learning (ML) framework for optimizing agricultural decision-making by integrating predictive modeling, clustering techniques, and genetic algorithms. This framework aims to enhance crop yield and net return predictions while considering environmental factors, resource constraints, and market dynamics in the Bargarh Canal command (BCC) of eastern India. Three state-of-the-art ML algorithms (Random Forest (RF), XGBoost, and Long Short-Term Memory (LSTM), were implemented and compared using a comprehensive set of input features, including environmental, agronomic, and management factors. The XGBoost model demonstrated superior performance, achieving an average R² of 0.87 and RMSE of 0.32 for yield prediction, and an R² of 0.83 with an RMSE of 0.52 tons/hectare for net return prediction across all crops. Four distinct crop clusters were identified, revealing trade-offs and opportunities for optimization. The framework enables various decision support applications, including crop calendar optimization, risk assessment, and resource allocation planning. Recommendations include weather-responsive planning, cluster-based diversification, precision agriculture investments, monitoring system enhancements, and policy interventions to promote sustainable agricultural intensification.
