Data-driven analysis and machine learning-based forecasting for big onion production in Sri Lanka

Abstract

The cultivation of big onions forms a critical part of Sri Lanka’s agriculture-based economy. However, the industry struggles with a challenge of supply and demand discrepancies, in addition to a high reliance on imports. This study proposes a machine learning-based predictive model to forecast production levels of big onions with a view to enhancing farmers’ and policymakers’ decisionmaking processes. Utilizing historical data from the Department of Census and Statistics and various institutions, a number of regression techniques including Linear Regression, Decision Tree, Random Forest, and Gradient Boosting were implemented to forecast production levels. The effectiveness of these methods was compared using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R2 score. It was revealed that with a set of defined hyperparameters, the Random Forest algorithm provides the best predictions with a score of 78.20 percent R2. The developed model has been deployed as a userfriendly web application using Streamlit and FastAPI, thus making it easier for stakeholders to use. This study forms a stepping stone to further application in using data-driven methods to enhance production forecasting in agriculture. Further studies may focus on expanding the dataset and using sophisticated deep learning methods to achieve even better results.