Machine Learning Approach to Measure the Impact of Physical Environment Features on Neighborhood Ambient Population Density

Abstract

The physical environment of a city and its ambient population density (the true density of people using the space) may be closely related. However, measuring these impacts directly is challenging because it involves many interacting features. This study proposes a way to evaluate these complex relationships by Machine Learning (ML). A tree-based classification model was trained using physical environment features of a neighborhood as input paired with ambient population density information represented by Mobile Spatial Statistics data. accuracy. Model interpretation tools such as SHAP and PDPs were then used to understand how the model makes decisions and to quantify the influence of each physical environment feature on ambient population density.