A Machine Learning-based Crop Yield Forecasting and Recommendation System

Abstract: Reliable agricultural forecasting is essential for effective crop planning, yield estimation, and climate-aware decision making, particularly in regions where farming activity is closely tied to seasonal rainfall. This paper presents an integrated machine learning-based agricultural prediction system that addresses three core tasks: crop selection, crop yield forecasting, and rainfall estimation using district-level datasets from Karnataka, India.

The proposed system employs optimized tree-based ensemble models, including Random Forest, XGBoost, LightGBM, and CatBoost, which are well suited for structured agricultural data commonly available in developing regions. Instead of relying on deep sequence models that require long temporal weather records and genotype-related inputs, the system operates on seasonal and regional attributes, enabling efficient training and deployment on modest hardware. The trained models are integrated into a web-based portal developed using PHP, Python, and MySQL, allowing farmers to access predictions through simple and intuitive interfaces. Comparative evaluation against a recent LSTM-based yield prediction framework shows that the proposed system achieves competitive or improved accuracy while remaining significantly more practical for real-world deployment.

Keywords: Agriculture, Machine Learning, Crop Prediction, Yield Forecasting, Rainfall Prediction, Web Application, Decision Support System.