Speaker
Description
The rapid increase in property transactions and rental prices in Taiwan has created serious challenges for housing equity, especially for college students without access to on-campus housing. Rising rents, combined with a lack of transparency and market information asymmetry, have exacerbated financial burdens and made the rental process increasingly stressful. The factors and spatial patterns that influence rental prices in college neighborhood are complex, and current research methods have shown limited capacity to explore the complexities. This study aims at geographic weighted machine learning (GW-ML) models to analyze rental price spatial patterns and their influencing factors for off-campus housing for college students.
The study begins by applying global Moran’s I and local LISA to identify high- and low-rent clustering areas. Subsequently, the K-means machine learning algorithm is used to classify rental prices into distinct tiers, capturing renter preferences and facilitating effective market segmentation. To account for spatial heterogeneity in off-campus rentals, the study employs GW-ML, a framework that integrates Geographic Weighted Regression (GWR) principles with advanced machine learning techniques.
Nine models are tested to evaluate their predictive performance, including linear regression models (Linear Regression, Bayesian Ridge, and Lasso Regression) and cutting-edge machine learning models such as Random Forest, XGBoost, Gradient Boosting Decision Tree (GBDT), LightGBM, Support Vector Regression (SVR), and k-Nearest Neighbors (KNN). Model performance is assessed using root mean squared error (RMSE). By incorporating spatial weights derived from kernel density functions, GW-ML captures the spatial and non-linear characteristics of rental price data more effectively than traditional ordinary least squares (OLS) housing price models.
Among the tested models, LightGBM demonstrates the highest predictive accuracy, effectively identifying spatial patterns and predicting rental price variations. Statistical analysis reveals that proximity to colleges, housing age, and access to transportation infrastructure (e.g., public bike systems and bus stops) are significant factors influencing off-campus rental prices. The findings highlight the potential of GW-ML to enhance understanding of spatial dynamics in rental markets and provide actionable insights for improving affordability and accessibility in college-area housing.
Future research could use graph neural networks and spatiotemporal analysis to better understand the dynamic interactions in rental preference. These approaches could help predict changes in rental prices over time and across locations, providing deeper insights into market trends and supporting better policy decisions.
| Keywords | geographically weighted machine learning; machine learning; housing rents |
|---|---|
| Best Congress Paper Award | Yes |
