Estimating Property Facility Scores Using Machine Learning: A Case Study Using Real Estate Data

Abstract

The assessment of interior facilities in condominiums is crucial as it serves as a significant information source for buyers and tenants, and it is also important for assessing the future profitability of real estate. In this study, we quantify the property management professionals' subjective perception of facilities and replicate subjective facility scores using machine learning. Initially, we interviewed professionals specializing in facility assessment to create a scoring sheet for facility, which was used to evaluate actual properties, forming our training dataset for modeling. Subsequently, we constructed a model using features from sales drawings data to estimate facility scores. Finally, we assessed how the model's accuracy was affected by missing data in sales drawings' features. Results indicated that consolidating expert knowledge into a scoring table enabled precise scoring that aligned with expert subjectivity, and the machine learning model achieved high accuracy with minimal variables. The model also demonstrated robustness in the presence of missing data in certain variables.