Integrating machine learning and discrete choice modeling for enhanced shopping destination choice model

Abstract

• Develops a modeling framework using machine learning and econometrics for parcel-level shopping destination choice. • Generates shopping location choice sets based on business types and locations. • Predicts individual shopping location choices using an econometric model considering unobserved behavior variations. • Results show longer travel times decrease routine shopping, while high retail areas attract shoppers variably. This study develops a two-stage modeling framework for parcel-level shopping destination choice, accounting for multi-dimensional factors and the heterogeneity in shopping location choice behavior. The study follows two steps: (i) developing a shopping location choice set generation process comprising feature selection and encompassing business types and locations, and (ii) developing an econometric model to predict individual shopping location choice behavior considering unobserved heterogeneity. The study advances a novel approach of combined machine learning (ML) and random utility-based discrete choice modeling (i.e., mixed logit model (MXL)). Results from the MXL model reveal that the longer the travel time and distance from the central business district, the less likely people are to visit a store for routine shopping (e.g., groceries). The random parameter analysis reveals that although high retail concentration surrounding the desired shopping location should attract individuals for shopping, there will be people who still may not intend to shop at those locations. Similarly, people may be willing to travel to stores requiring longer travel times for special item shopping. The models developed in this study will be implemented within an integrated transport, land use, and energy (iTLE) modeling system to improve the behavioral representation of destination choices.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 2 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.