Segmenting U.S. households by behavioral patterns to predict food waste: A data-driven approach using public datasets

Household food waste constitutes a significant sustainability and food security challenge in the United States, with substantial environmental and social implications. This study integrates publicly available datasets – including the USDA’s Food Acquisition and Purchase Survey (FoodAPS 2012–2013), the 2018 American Community Survey, and ReFED’s 2018 regional food waste estimates – with machine learning techniques to model and predict household-level food waste. We applied regression, classification, and clustering approaches to analyze waste behaviors. Among the predictive models tested, a random forest regression provided the most accurate predictions of household food waste, outperforming other methods. Classification models were used to predict Supplemental Nutrition Assistance Program (SNAP) participation and to assign households to waste-behavior clusters identified by unsupervised learning. Demographic factors, particularly household size and poverty ratio were among the strongest predictors of household food waste, while behavioral indicators such as grocery list frequency and food sufficiency played a secondary role. Clustering revealed distinct household profiles with varying waste patterns, differentiating, for example, food-insecure SNAP-dependent families from larger resource-stable households. To ensure alignment with real-world waste quantities, model outputs were calibrated against ReFED’s regional waste data. The findings demonstrate the value of integrating diverse public datasets with machine learning to uncover drivers of household food waste. These insights enable more targeted household-level waste reduction interventions and support the development of effective, data-driven policies for food waste mitigation.