Remote sensing-based house value estimation using an optimized regional regression model

Abstract

This study proposed a new method to predict residential property value using remote sensing data as a major data source substitute to traditional inputs in house price estimation models. An optimized regional regression (ORR) approach was proposed in this study. This approach integrated a differential evolution optimization algorithm along with the ordinary least square regression to improve house value prediction accuracy. In addition to ORR, four other regression methods, random forest, Cubist regression trees, geographically weighted regression, and global ordinary least square, were also employed to provide a comparison. Results showed that models using remote sensing data are capable of acquiring accurate house price information. In addition, the volume of residential buildings proved to be an efficient substitute for total living area, the most important variable of the house price estimation model (i.e., a hedonic model). The ORR approach yielded the most accurate predictions followed by the geographically weighted regression. Further investigation indicated that the ORR approach has three major advantages: it is effective, stable, and the results are readily interpretable.