Urban Commercial Site Selection and Station Passenger Flow Prediction Based on Machine Learning: A Case Study of Tokyo’s Yamanote Line

Abstract

This study investigates the application of machine learning techniques to analyze passenger flow data and predict optimal commercial store types at stations along Tokyo’s Yamanote Line. The Yamanote Line connects major commercial districts and bustling urban areas, with high passenger volumes and diverse consumer activity, making it a prime region for strategic retail deployment. However, determining the optimal store type and location remains a major challenge for businesses due to limited data-driven tools.
To address this, we propose a predictive framework that integrates open datasets from the Japanese Statistics Bureau, Seikatsu Guide.com and e-Stat. These datasets include station-level traffic, census demographics, consumer expenditure, and geographic information within a one-kilometer radius of each station. Feature selection was conducted using the Random Forest algorithm to identify key determinants influencing commercial placement. Subsequently, multiple classifiers—including Gradient Boosting, Logistic Regression, Random Forest, and Extreme Gradient Boosting (XGBoost)—were evaluated for prediction performance.
Experimental results show that XGBoost outperformed other models, achieving an F1-score of 0.642 and accuracy of 0.68. By constructing a data-driven location intelligence model, this research supports evidence-based commercial site selection and enhances decision-making in competitive urban retail environments. The proposed framework not only improves prediction accuracy but also offers practical guidance for retail site planning in smart cities globally. Future research could explore the incorporation of real-time dynamic data and deep learning techniques to further improve accuracy and adaptability.