食品ロス削減に向けた食品の発注精度向上に関する研究

抄録

In this study, multiple regression analyses on food sales were made by statistical and machine learning models for the shops at the Kitakata Campus and the Hibikino Campus of the University of Kitakyushu. We aim to establish a prediction model of sale amount that contributes to the improvement and reduction of food loss. The data on daily sale amounts of foods with short expiration dates (bread, boxed lunches, and rice balls) from November 2021 to November 2022 were provided by University Corporation shops. Through analysis, we found that there was a good correlation between the daily sale amounts of food and four factors e.g., yearly-averaged daily sales by weekday, sale amounts on the same day of last week, attending school or not, and weather conditions. Moreover, we developed a detailed future prediction model using multiple regression analysis and machine learning techniques and validated their accuracies by using the indicator of Root Mean Square Error. After comparison, we found that the predictions for two campuses made by the machine learning-based models were a little weaker than those made by multiple linear regression in this case, which was caused by the relatively small data volume. Moreover, it was demonstrated that the optimal feature (factor) selection rather than the number of features was essential for improving the prediction accuracy.