Design and Implementation of a Machine Learning Simulation System for Crop Yield Prediction Under Climate Variability

Climate variability continues to threaten agricultural productivity, especially in regions where farming depends heavily on rainfall and temperature conditions. This study presents the design and implementation of a web-based computational simulation system for predicting crop yield under climate variability. The system uses rainfall, temperature, pesticide usage, crop type, country, and land size as input variables to estimate crop yield and total production. Historical crop yield and climate data were processed, filtered, and used to train crop-specific machine learning models. Feature engineering was applied to generate additional predictors, including temperature-rainfall interaction, pesticide intensity, growing degree days, heat stress, cold stress, rainfall categories, logarithmic variables, and squared variables. Random Forest, Gradient Boosting, Ridge Regression, and ensemble models were trained and evaluated. The system was implemented using a Fast API backend, a web-based frontend, interactive visualization components, and an AI-based recommendation module that explains completed prediction results. The results showed strong predictive performance for several crops, with rice achieving the highest test R² value of 0.9617, followed by maize with 0.9080, soybeans with 0.8524, wheat with 0.8157, and potatoes with 0.7841. The study demonstrates that machine learning-based simulation can support crop yield prediction and agricultural decision-making under changing climate conditions, especially in data-limited regions.