Crop Yield Prediction Based on Indian Agriculture Using Machine Learning

Abstract

This paper develops a machine-learning framework to estimate crop yield per hectare in different Indian districts based on open-agricultural data. The system combines data preprocessing, correlation-based and LASSO feature selection methods, and the performance evaluation of six regression algorithms: Ridge, Decision Tree, Random Forest, XGBoost, LightGBM, and CatBoost. The CatBoost model was able to perform well with an R2 close to 0.93 and a low RMSE, demonstrating accurate yield prediction can be achieved using agro-economic and spatial features without explicit soil and weather data. Furthermore, the model is made available through a FastAPI backend along with a React-based multilingual web interface, which collectively offers live, easy-to-reach, and data-driven insights for farmers and policymakers.