Leveraging machine learning to predict residential location choice: A comparative analysis

Residential location choice is a pivotal factor shaping urban landscapes, influencing economic activity, social demographics, and transportation networks. Understanding these decisions is essential for effective urban planning. While discrete choice models have been used to predict residential locations, their limitations in handling complex datasets have hindered their accuracy. Machine learning has shown potential to outperform discrete choice methods in transportation-related decisions like travel mode. However, the traditional application of these methods to residential location choice faces challenges due to the large number of choice alternatives and the limitations of sample size and data availability. This often leads to biased predictions, favoring popular choices over less common ones. This study introduces a novel approach that overcomes these limitations, making machine learning more effective for residential location choice. By applying this method and different machine learning models, the study provides a detailed comparison of their performance in predicting residential choices. A comparative analysis of various machine learning algorithms reveals that XGBoost and gradient boosting models significantly outperform traditional methods, achieving a 42 % accuracy rate in predicting residential location choices on the 33 % validation data of household travel survey data from MWCOG. These findings offer valuable insights for policymakers and urban planners to optimize land use, transportation, and urban development strategies.