Tracking of Doubtful Real Estate Transactions by Outlier Detection Methods: a Comparative Study

Doubtful real estate transactions, with the prices far away from the market prices, appear because of non commercial transactions or efforts in order to hide the taxes. To estimate the right values of parameters, such data must be removed from a data set or robust methods of parameters estimation are to be used, while developing a mass appraisal model. Such transactions are outlying observations, which can be detected and removed by outlier detection methods. The purpose of the work is to review outlier detection methods and to test the possibility of using them to solve the task. An overview of real estate market value, valuation methods and process of mass appraisal is made to introduce to real estate mass valuation. Overview of outlier detection method contains scaling and such methods: resampling by half means, the smallest half volume, the closest distance to the center, ellipsoidal multivariate trimming, minimum volume ellipsoid, minimum scatter determinant, analysis of projection matrix, principal components and residuals, also influence measures, robust regression, and classification methods. The reviewed methods were categorized; commonly used methods were selected and tested experimentally aiming to compare the effectiveness. Best results were achieved using the multilayer perceptron and the principal component analysis based technique.