Clustering Analysis with Embedding Vectors: An Application to Real Estate Market Delineation

Although clustering analysis is a popular tool in unsupervised learning, it is inefficient for the datasets dominated by categorical variables, e.g., real estate datasets. To apply clustering analysis to real estate datasets, this study proposes an entity embedding approach that transforms categorical variables into vector representations. Three variants of a clustering algorithm, i.e., the clustering based on the traditional Euclidean distance, the Gower distance, and the embedding vectors, are applied to the land sales records to delineate the real estate market in Gwacheon-si, Gyeonggi province, South Korea. Then, the relevance of the resultant submarkets is evaluated using the root mean squared errors (RMSE) obtained from a hedonic pricing model. The results show that the RMSE in the embedding vector-based algorithm decreases substantially from 0.076-0.077 to 0.069. This study shows that the clustering algorithm empowered by embedding vectors outperforms the conventional algorithms, thereby enhancing the relevance of the delineated submarkets.