Predictive Regression Model Evaluation : Evaluating Predictive Machine Learning Models to Reduce Food Waste in the Dairy Industry

Abstract: Food waste in the food industry is often a result from the complex nature of food production. Uncertainty is always present as yields vary and as there is a chain of consumer demand from stores to producers. Food waste is a waste of both natural and economic resources affecting both the planet and the producer. The environmental impact is further affected as excessive production that leads to waste correlates to an excessive carbon footprint from excessive farming, transports, and processing. Additional environmental impacts are excessive land usage and overfertilization. In order to aid in the reduction of food waste in the food industry, this thesis evaluates a machine learning approach to predicting commercial waste. This thesis evaluates 11 predictive machine learning models on their predictability on commercial food waste at a large product producing company in the dairy industry. The models are evaluated according to the two metrics; Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). Based on the models’ RMSE and MAE metrics, the Random Forest model proved to be the most suitable model. It presented the best and statistically significant RMSE. It also produced the best MAE but without statistical significance to gradient boost and Support Vector Regression (SVR), indicating that the difference in performance between the models, according to MAE, are within a variation span that can occur due to chance.