Machine Learning-Assisted Dynamic Proximity-Driven Sorting Algorithm for Supermarket Navigation Optimization: A Simulation-Based Validation

Abstract

In-store grocery shopping is still widely preferred by consumers despite the rising popularity of online grocery shopping. Moreover, hardware-based in-store navigation systems and shopping list applications such as Walmart’s Store Map, Kroger’s Kroger Edge, and Amazon Go have been developed by supermarkets to address the inefficiencies in shopping. But even so, the current systems’ cost-effectiveness, optimization capability, and scalability are still an issue. In order to address the existing problems, this study investigates the optimization of grocery shopping by proposing a proximity-driven dynamic sorting algorithm with the assistance of machine learning. This research method provides us with an analysis of the impact and effectiveness of the two machine learning models or ML-DProSA variants—agglomerative hierarchical and affinity propagation clustering algorithms—in different setups and configurations on the performance of the grocery shoppers in a simulation environment patterned from the actual supermarket. The unique shopping patterns of a grocery shopper and the proximity of items based on timestamps are utilized in sorting grocery items, consequently reducing the distance traveled. Our findings reveal that both algorithms reduce dwell times for grocery shoppers compared to having an unsorted grocery shopping list. Ultimately, this research with the ML-DProSA’s optimization capabilities aims to be the foundation in providing a mobile application for grocery shopping in any grocery stores.