Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FoodRL: A Reinforcement Learning Ensembling Framework For In-Kind Food Donation Forecasting

Esha Sharma, Lauren Davis, Julie Ivy, Min Chi

TL;DR

FoodRL introduces a reinforcement learning-based metalearning framework that enforces clustering to reduce action space and dynamically weight ensembles of diverse time-series forecasts for in-kind food donations. By modeling ensembling as a sequential decision problem and constraining the action space with model clustering, FoodRL achieves superior MAPE/MAE across two U.S. food banks facing concept drift from disasters and seasonal fluctuations. The approach outperforms simple averaging, genetic algorithms, and vanilla RL, providing robust performance under drift and offering a path to real-time redistribution planning with potential social impact in the millions of meals per year. Practical deployment is underway with industry partners, and future work will extend performance during extreme events and weekly forecasts, while generalizing the framework to other humanitarian logistics domains.

Abstract

Food banks are crucial for alleviating food insecurity, but their effectiveness hinges on accurately forecasting highly volatile in-kind donations to ensure equitable and efficient resource distribution. Traditional forecasting models often fail to maintain consistent accuracy due to unpredictable fluctuations and concept drift driven by seasonal variations and natural disasters such as hurricanes in the Southeastern U.S. and wildfires in the West Coast. To address these challenges, we propose FoodRL, a novel reinforcement learning (RL) based metalearning framework that clusters and dynamically weights diverse forecasting models based on recent performance and contextual information. Evaluated on multi-year data from two structurally distinct U.S. food banks-one large regional West Coast food bank affected by wildfires and another state-level East Coast food bank consistently impacted by hurricanes, FoodRL consistently outperforms baseline methods, particularly during periods of disruption or decline. By delivering more reliable and adaptive forecasts, FoodRL can facilitate the redistribution of food equivalent to 1.7 million additional meals annually, demonstrating its significant potential for social impact as well as adaptive ensemble learning for humanitarian supply chains.

FoodRL: A Reinforcement Learning Ensembling Framework For In-Kind Food Donation Forecasting

TL;DR

FoodRL introduces a reinforcement learning-based metalearning framework that enforces clustering to reduce action space and dynamically weight ensembles of diverse time-series forecasts for in-kind food donations. By modeling ensembling as a sequential decision problem and constraining the action space with model clustering, FoodRL achieves superior MAPE/MAE across two U.S. food banks facing concept drift from disasters and seasonal fluctuations. The approach outperforms simple averaging, genetic algorithms, and vanilla RL, providing robust performance under drift and offering a path to real-time redistribution planning with potential social impact in the millions of meals per year. Practical deployment is underway with industry partners, and future work will extend performance during extreme events and weekly forecasts, while generalizing the framework to other humanitarian logistics domains.

Abstract

Food banks are crucial for alleviating food insecurity, but their effectiveness hinges on accurately forecasting highly volatile in-kind donations to ensure equitable and efficient resource distribution. Traditional forecasting models often fail to maintain consistent accuracy due to unpredictable fluctuations and concept drift driven by seasonal variations and natural disasters such as hurricanes in the Southeastern U.S. and wildfires in the West Coast. To address these challenges, we propose FoodRL, a novel reinforcement learning (RL) based metalearning framework that clusters and dynamically weights diverse forecasting models based on recent performance and contextual information. Evaluated on multi-year data from two structurally distinct U.S. food banks-one large regional West Coast food bank affected by wildfires and another state-level East Coast food bank consistently impacted by hurricanes, FoodRL consistently outperforms baseline methods, particularly during periods of disruption or decline. By delivering more reliable and adaptive forecasts, FoodRL can facilitate the redistribution of food equivalent to 1.7 million additional meals annually, demonstrating its significant potential for social impact as well as adaptive ensemble learning for humanitarian supply chains.

Paper Structure

This paper contains 31 sections, 13 equations, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Prior Work on Food Donation Forecasting
  4. The Necessity of Ensembling Learners to Address Concept Drift
  5. Metalearning for Time Series Models
  6. Simple Averaging
  7. Genetic algorithm
  8. Reinforcement Learning Ensembling Approach
  9. Methods
  10. State
  11. Action
  12. Reward
  13. Transition Dynamics.
  14. Objective.
  15. Experimental Setup
  16. ...and 16 more sections

Figures (1)

  • Figure 1: Concept Drift in Food Bank Data