Analyzing the Impact of Land-Use Characteristics and Demographic Factors on Spatial Variations in Public Bus Usage: A Comparison of Pre- and During COVID-19 Periods

Abstract

The spread of the coronavirus pandemic led to significant changes in bus-usage patterns in urban areas worldwide. Researchers have frequently employed linear and nonlinear models in bus-usage studies. However, existing linear models assume that each variable affects a uniform range, limiting their ability to capture localized pattern changes. This study applies a multiscale geographically weighted regression model reflecting the characteristics of the variables to address these limitations. Linear models are constrained by their inability to account adequately for the complex dynamics of real-world bus usage. This research introduces nonlinear methods to overcome these constraints. The geographical random forest method, an advanced variant of the random forest model, integrates spatial concepts to explain local patterns more effectively than traditional machine learning techniques. The linear models revealed significant changes in four variables (i.e., population size, over-65 population ratio, number of students, and land-use complexity). In contrast, nonlinear models demonstrated diverse movement patterns influenced by several factors, indicating a shift toward new public transportation patterns.