Estimating available supermarket commodities for food bank collection in the absence of information

Food insecurity is a widespread concern in the United States. Addressing this concern is a chief goal of many non-profit organizations including food banks. Understanding the availability of donations is beneficial when addressing the demand in local communities, especially when their collection requires food bank-managed vehicles. High-volume donors such as supermarkets, however, do not provide information in regards to what items are available. This can negatively impact inventory management capabilities and cause unnecessary transportation costs. This research evaluates four approximation methods based on their ability to estimate food availability at supermarkets including the multiple layer perceptron artificial neural network, multiple linear regression, and two naïve estimates for the average collection amount. Using a subset of the historic data provided by the Food Bank of Central and Eastern North Carolina (FBCENC), the four approximation methods are evaluated in terms of their ability to estimate collection amounts in the next planning period. Transportation cost estimates are then calculated using projections made using each approximation method and compared to those calculated using the actual transportation costs. Results suggest that the MLP-NN models provide the best approximations for each food type and provide closer estimations for transportation cost than other approximation methods.