Food Purchase Prediction Using Multi-Algorithmic Approaches Enhanced by A Stacking Ensemble Classifier

Abstract

In the digital era, online food delivery platforms such as Zomato and Swiggy have significantly transformed customer purchasing behavior, shifting from traditional dining to online ordering. This evolution presents challenges in accurately predicting customer preferences and ordering patterns, which are essential for effective resource management and personalized services. To address this, a multi-algorithmic approach using a stacking ensemble classifier is proposed for food purchase prediction. The model integrates Random Forest, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Logistic Regression as base learners, with Random Forest as the meta-classifier. It utilizes demographic and behavioral features, including age, gender, income, relationship status, and past feedback. Experimental results demonstrate that the proposed stacking model achieves superior performance with an accuracy of 93.17%, outperforming individual classifiers. The findings highlight the effectiveness of ensemble learning in enhancing predictive accuracy and supporting decision-making in online food delivery systems.