Title: Least median of squares regression and minimum volume ellipsoid estimator for outliers detection in housing appraisal

Authors: Pierluigi Morano; Francesco Tajani

Addresses: Department of Science of Civil Engineering and Architecture, Polytecnic of Bari, Via Orabona 4, Italy ' Department of Civil Engineering, University of Salerno, Via Giovanni Paolo II 132, Italy

Abstract: In the real estate sector, regression analysis is the most used method for interpretative and predictive purposes. However, the presence of outliers in the estimative sample can lead to ordinary last squared regression models that do not represent the investigated market phenomenon, with the consequence of producing unreliable assessments. In the present research, the issues of the identification and the removal of outliers are discussed. The outliers identified by the least median of squares (LMS) regression and the minimum volume ellipsoid (MVE) estimator are compared in order to test the coincidence or the diversity. A complete diagnosis of the data of the initial estimative sample is carried out, combining the robust residuals obtained with LMS and the robust distances obtained with MVE. The data are classified into regular observations, vertical outliers, good leverage points and bad leverage points, and cases to delete and those to keep in the sample are identified.

Keywords: real estate appraisal; outliers; least median of squares regression; minimum volume ellipsoid estimator; outlier detection; housing appraisal.

DOI: 10.1504/IJBIDM.2014.065074

International Journal of Business Intelligence and Data Mining, 2014 Vol.9 No.2, pp.91 - 111

Published online: 24 Oct 2014 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article