Automated Real Estate Appraisal Methods - A comparison of multiple linear regression and artificial neural networks in real estate valuation

Abstract

The field of real estate appraisal is in the midst of a rapid normalization of automated mass appraisal methods. Automated mass appraisal methods are computer assisted data driven appraisal methods. These appraisal methods include multiple linear regression and artificial neural networks that both fall under the umbrella term machine learning. New companies and services are built based on the use of these mass appraisal techniques, raising the importance of proper use of them and knowledge of the possible pros and cons.

This thesis compares the multiple linear regression and artificial neural networks methods on their appraisal performance on the sales data available for individuals of the Finnish city of Tampere. The motivation for this study is to better understand the possible performance differences of these two methods and see how well they can be implemented with the limited data available for the public. For reasonable comparison, several performance metrics were used along with data engineering and statistical techniques when deemed appropriate.

The analysis results align somewhat with the previous literature of the performance of the two methods. The performance of the artificial neural network method is found to be significantly better than the performance of multiple linear regression when measured by mean squared error, mean absolute percentage error and R2. However, the performance difference could partially be due to the choice in the data engineering part as previous studies report performance advantages to be tied to dataset size and data engineering choices.