● Food donation prediction models remain scarce for New York State. ● GBRT and XGBoost outperformed RF for donation prediction. ● Historical donations and donor proximity were key predictors. ● Supply–demand mismatch highlights the needs of interventions. Food donation reduces food waste and alleviates hunger, ranking as the second most preferred strategy in the food waste management scale suggested by the US Environmental Protection Agency. Food donation prediction models are critical for food donation planning, food waste management, and achieving sustainability goals of zero hunger, resource efficiency and environmental protection. However, food donation prediction models are limited. These existing models are based on limited data from three non-New York states and fail to account for New York’s unique challenges where 4M tons of annual food waste (17% of municipal waste) coexists with 24.9% adult food insecurity. To address these knowledge gaps, this study employed three advanced machine learning techniques—Random Forest, Gradient Boosted Regression Trees, and Extreme Gradient Boosting—to predict food donation amounts using data from 233 retail stores in New York State. Our model comparison revealed that Gradient Boosted Regression Trees model was the best-fit model. Our findings highlight that the historical donation amounts and donor proximity to food banks as critical factors influencing donation amounts, indicating food donations are primarily driven by donors' strategic, moral, and economic motives rather than recipients' needs. Finally, SNAP food stamp usage and median household income are not key predictors of food donation, which may suggest a potential mismatch between supply and demand, emphasizing the need for more sustainable and equitable food redistribution networks.
Romeiko et al. (Fri,) studied this question.