Predicting food leftovers in shared dining: a machine learning approach

Purpose Existing literature in the hospitality industry predominantly addresses food waste after it occurs, leaving a gap in predictive knowledge. This research aims to fill this gap by developing a prediction model focused on shared dining, which often results in a large quantity of food leftovers because of discrepancies in diners’ expectations and over-ordering to satisfy everyone. Design/methodology/approach The prediction model was validated in two stages. First, guided by theoretical foundations and empirical evidence, nearly half a year of extensive observations at a dim sum restaurant yielded 1,784 transactional data points and corresponding food leftover records. Second, multiple machine learning models were built and compared to identify the most parsimonious model that effectively predicts food leftovers. Findings The prediction model identified 11 positive and 4 negative factors as the most effective configuration for predicting food leftovers. Key predictors included demographic elements such as age and group size, along with behavioral factors like ordering patterns and meal preferences. These findings underscore the complexity of food waste behaviors in shared dining settings (i.e., dim sum dining) and highlight the importance of a multifaceted predictive approach. Originality/value The developed prediction model contributes theoretically and practically to the hospitality industry. Theoretically, it enhances the understanding of the factors influencing food leftovers in shared dining contexts. Practically, it provides hospitality practitioners actionable insights for implementing preventive measures, promoting sustainable dining management, and reducing food waste.