In-Store Traffic Density Estimation

Estimating in-store traffic density is critical for a product allocation that improves the retailer profit and enhances the customer shopping experience. We propose a way to estimate traffic densities based on a regression analysis carried over e-receipts from a supermarket in Lebanon. Our approach simplifies alternative approaches in the specification of the regression variables. We obtain the dependent variable (i.e. the traffic) via a demand filtering approach without reconstructing shopping paths. We then estimate the independent variable reflecting the way product allocation drives traffic in neighboring shelves through a straightforward attraction approach. We propose and compare alternate regression models, such as support vector regression, regression trees, kernel regression, and Gaussian processes. Our results show that regression trees and kernel regression methods perform well.