An Intelligent Platform For Surplus Food Management Using Machine Learning

Abstract

Food waste and hunger together represent a critical global challenge, with vast quantities of edible food being discarded while millions of people continue to face food insecurity. There is an urgent need for efficient and inclusive systems to redistribute surplus food to those in need. Existing solutions rely heavily on manual coordination, which significantly limits their effectiveness, especially in developing regions and among underserved populations. To address this challenge, the project proposes an intelligent online public food-sharing platform that integrates machine learning with rule-based decision systems and accessible infrastructure. The system is hybrid in nature, combining data-driven learning models with deterministic rule-based logic interpretability. Food donors submit entries via mobile applications or public display units, which are then uploaded to a locally hosted backend server for processing. A Logistic Regression model is used to filter out spam and ensure the authenticity of listings, thereby improving platform reliability. The model calculates the probability of a listing being valid based on features such as user behavior, content, and posting patterns. Its performance is evaluated using a confusion matrix, which helps assess how accurately the model distinguishes between spam and genuine listings. Additionally, bias-variance analysis is conducted to determine whether the model is underfitting or overfitting. A Random Forest model is used to prioritize food postings based on pickup likelihood, considering factors such as location, time, and historical patterns. As an ensemble learning method, Random Forest improves prediction accuracy by combining multiple decision trees and reducing individual model errors. Additionally, a rule-based expiry model estimates food freshness and shelf life using parameters such as food type, preparation time, and storage duration, ensuring safe and timely distribution. The Logistic Regression model demonstrates stable and balanced performance with low variance, making it suitable for reliable classification tasks. The Random Forest model achieves higher accuracy, as reflected in its confusion matrix, by reducing prediction errors through ensemble voting. Furthermore, the learning curve indicates improved model performance with an increase in training data size. The integration of public display units in high-traffic areas is designed to reduce dependency on personal digital devices and improve accessibility for a broader population. Overall, this project demonstrates the potential to enhance surplus food utilization and reduce waste through intelligent, data driven decision-making. Use technical words and regenerate.