Determining the Flat Sales Prices by Flat Characteristics Using Bayesian Network Models

Abstract

There are various factors affecting flat sales prices. Various characteristics of a flat play an important role in determining its sales price. In this study, a machine learning based Bayesian network was built by a restrictive structural learning algorithm using the data collected from 24 randomly selected cities in Turkey. The data consist of the sales prices and various characteristics of a flat such as number of bedrooms, building age, availability of balcony, net area, heating type, mortgageability, number of bathrooms, seller type, presence in a housing estate area and floor location. After the model validity check, a sensitivity analysis was performed for the estimated Bayesian network model and related results were provided. Some of these results indicate that sales prices of flats mostly change depending on the number of bathrooms available. Additionally, number of bedrooms, net area and floor location are also determinative about the sales prices. The lack of significant difference among the sales prices of flats that are sold by construction companies, housing estate agents or property owners is another result obtained.