In many countries, and especially in regions where agriculture is the predominant source of income and food, agriculture is of major economic importance and is a vital pillar of the country. Therefore, accurate estimation of yield is essential for resource planning and management, policy development, and optimization of the goods movement chain The present study investigates in detail the extent to which several machine learning, especially non-linear, models such as Random Forest, XGBoost, Decision Tree Regressor, Gradient Boosting Machines, CatBoost, and Support Vector Regression (SVR) can help in forecasting crop yield of economically important crops like rice, sorghum, and cotton grown in various states of India more efficiently. The dataset used in this study spans from 1997 to 2019 and includes agronomic data, environmental data, and time series data. Models were trained on data from 1997–2018, with 2019 data set aside for out-of-sample forecasting validation. Performance was assessed by R2, RMSE, and MAE metrics. The analysis revealed that the Random Forest model recorded the best results, achieving an R2 score of 0.9865, with the lowest RMSE and MAE among all models tested, demonstrating its superior accuracy in predicting crop yields. Ensemble models such as XGBoost and Gradient Boosting Machines also performed competitively, further confirming the value of advanced ML algorithms in capturing complex agricultural data patterns. This study underscores the significance of modern ML algorithms in developing models that enhance agricultural practices and support sustainable agricultural decision-making.