Application of machine learning in predicting fruit waste in a South African fresh produce wholesale market

Machine learning has been generally used for prediction and classification tasks in the food value chain. However, its application in the study of food waste has been limited. Therefore, this study explored the potential of predicting fruit waste at a wholesale level of the food value chain, using a fresh produce wholesale market in South Africa as a case study. The study aimed to develop a machine learning model to predict fruit waste during marketing. Using historical data at the case study market from 2021 to 2023, different machine learning algorithms such as Random Forest, Gradient boosting, Decision tree, XGBoost, Extra tree and a Stacked Model were applied. The results revealed that fruits in the category of melons and citrus contributed more to fruit waste at the market, while the most waste was during spring and summer seasons, with the highest waste occurring in 2022. The decision tree and extra tree models were the most promising among the machine learning models in the training dataset, with an MAE of 112.19 each. At the same time, the XGBoost outperformed other models for the testing dataset with an MAE of 232.32. The study provided a solid baseline for future studies in this area and recommended integrating varied data for a more robust and accurate model. With further research and implementation, the developed machine learning model has the potential to aid market decisions and policymaking to reduce postharvest waste of fruits at the market, thereby enhancing profitability and sustainability.