Journal of Housing and the Built Environment ( IF 2.033 ) Pub Date : 2021-03-02 , DOI: 10.1007/s10901-021-09824-1 Jieh-Haur Chen , Tingting Ji , Mu-Chun Su , Hsi-Hsien Wei , Vidya Trisandini Azzizi , Shu-Chien Hsu
Data-driven housing-market segmentation has been given increasing prominence for its objectiveness in identifying submarkets based on the housing data’s underlying structures. However, when handling high-dimensionality housing dataset, traditional statistical-clustering methods have been found to tend to lose low-variance information of the dataset and be deficient in deriving the globally optimal number of submarkets. Accordingly, with the intention of achieving more rigorous high-dimensionality housing market segmentation, a swarm-inspired projection (SIP) algorithm is introduced by this study. Using a high-dimensionality Taipei city’s housing dataset in a case study, a comparison of the proposed SIP algorithm and a statistical-clustering method using the combination of principal component analysis (PCA) and K-means clustering is conducted in evaluating the predictive accuracy of hedonic price models of the housing submarkets. The results show that, as compared to the original single market, the segmented submarkets resulting from SIP algorithm are more homogenous and distinctive, where the resulted hedonic price models have high-level statistical explanation and disparate sets of hedonic prices for different submarkets. In addition, as compared to the use of a statistical-clustering method, SIP algorithm is found to obtain a more optimal number of submarkets, where the resulted hedonic price models are found to achieve greater improvement of statistical explanation and more stable reduction of prediction error. These findings highlight the advantages of our proposed SIP algorithm in high-dimensionality housing market segmentation, and thus it is hoped that the present research will serve as a practical tool to better inform further studies aimed at market-segmentation-related problems.
中文翻译:
群体启发式数据驱动的住房市场细分方法:以台北市为例
数据驱动的住房市场细分在基于住房数据的基础结构确定子市场的客观性方面越来越受到重视。但是,在处理高维住房数据集时,发现传统的统计聚类方法往往会丢失数据集的低方差信息,并且在推导全球最优子市场数量方面存在不足。因此,为了实现更严格的高维度住房市场细分,本研究引入了群体启发式投影(SIP)算法。在案例研究中使用台北市的高维住房数据集,在评估住房子市场享乐价格模型的预测准确性时,对建议的SIP算法与使用主成分分析(PCA)和K-均值聚类相结合的统计聚类方法进行了比较。结果表明,与原始单一市场相比,由SIP算法产生的细分子市场更加同质和鲜明,其中享乐价格模型具有高层次的统计解释和针对不同子市场的享乐价格的不同集合。另外,与使用统计聚类方法相比,发现SIP算法可获取更多最佳子市场,发现享乐价格模型可以更好地改善统计解释并更稳定地减少预测误差。这些发现凸显了我们提出的SIP算法在高维住房市场细分中的优势,因此希望本研究将成为一种实用工具,以更好地为针对与市场细分相关问题的进一步研究提供信息。