Abstract

Crop yield prediction is one of the crucial steps in food safety. Prediction with low error rates can support optimizing resources and facilitate informed decision making. Machine learning (ML) has been proven as useful tool for processing large agricultural datasets to forecast and predict the crop yield and other aspects of agriculture. ML models can facilitate informed decision-making and improve agricultural productivity. There have been no significant studies available on Jordan's data; this study utilizes the agricultural dataset from the Department of Statistics in Jordan and the climate dataset from the Climate Change Knowledge Portal to understand and build machine learning models that predict crop yield. This study aims to train and evaluate machine learning models, including regression, time series methods, Auto-Regressive Integrated Moving Average (ARIMA), and the Holt-Winters model for forecasting agricultural yields. Model’s performance was evaluated with error metrics such as Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), R-squared (R2), and Adjusted R-squared, etc. Results indicate Lasso regression and Random Forest are the most effective models with the lowest error rates and highest R-squared values. Lasso regression gets a R-squared value of 0.99 with an MAE of 0.104, while Random Forest achieves an R-squared value of 0.99 with an MAE of 0.108 across all evaluation metrics.