Extraction of Characteristic Subgraph Patterns with Support Threshold from Databases of Floor Plans

Abstract

The rental price of apartments is determined by various factors such as building age, floor space, and so forth. Although room layouts, or connection structures among rooms, are considered to affect the price, it is not obvious that what kind of layout has a strong influence. Identification of characteristic layouts on the price has a large impact on property lenders to determine appropriate rental price as well as on property borrowers to judge the reasonableness. In this paper, a framework to extract characteristic patterns on room layouts is proposed based on advanced techniques in frequent subgraph mining. Floor plans are represented as graphs, and class labels are given based on the estimated residuals by certain basic attributes. For the set of labeled floor plans, we develop a new class of contrast pattern having support threshold, under new definitions of support value of a pattern in a floor plan. In addition, we extract decision lists and decision sets consisting of significant subgraph patterns with thresholds by using methods in interpretability research of machine learning. The effectiveness of the proposed framework is evaluated in preliminary experiments using a real world dataset.