Investigation of Incremental Support Vector Regression Applied to Real Estate Appraisal

Incremental support vector regression algorithms (SVR) and sequential minimal optimization algorithms (SMO) for regression were implemented. Intensive experiments to compare predictive accuracy of the algorithms with different kernel functions over several datasets taken from a cadastral system were conducted in offline scenario. The statistical analysis of experimental output was made employing the nonparametric methodology designed especially for multiple NxN comparisons of N algorithms over N datasets including Friedman tests followed by Nemenyi's, Holm's, Shaffer's, and Bergmann-Hommel's post-hoc procedures. The results of experiments showed that most of SVR algorithms outperformed significantly a pairwise comparison method used by the experts to estimate the values of residential premises over all datasets. Moreover, no statistically significant differences between incremental SVR and non-incremental SMO algorithms were observed using our stationary cadastral datasets. The results open the opportunity of further research into the application of incremental SVR algorithms to predict from a data stream of real estate sales/purchase transactions.