An Exploration of Neural Networks and Its Application to Real Estate Valuation

TLDR

The study investigates the use of neural networks for residential property appraisal by comparing two NN models to a traditional multiple regression approach. It analyzes 288 Fort Collins home sales, applying two distinct NN architectures and a regression baseline to estimate sales prices. The results show that neural networks do not outperform regression, exhibiting inconsistent outputs across packages and runs and long runtimes, so appraisers should use them cautiously.

Abstract

This research applies neural network (NN) technology to real estate appraisal and compares the performance of two NN models in estimating the sales price of residential properties with a traditional multiple regression model. The study is based on 288 sales of homes in Fort Collins, Colorado. Results do not support previous findings that NNs are a superior tool for appraisal analysis. Furthermore, significant problems were encountered with the NN models: inconsistent results between packages, inconsistent results between runs of the same package and long run times. Any appraiser who plans on using this new technology should do so with caution.