Applying text mining methods to classify maintenance conditions for Real Estate Valuation

This study describes the potential of applying text mining methods for estimating the state of maintenance of real estate through advertisement texts. In order to succeed in the current growing housing market, mass appraisals are applying real estate valuation models on a large scale. However, these models are still striving for improvements in terms of accuracy, explainability and objectivity. Improving the classification of maintenance conditions by analyzing real estate advertisement through text mining methodologies could contribute to these estimation improvements. Firstly, by implementing a dataset of approximately 65,000 real estate samples and combining count-based and word embeddings text mining methodologies with supervised machine learning classification algorithms, we created twelve models suitable for predicting maintenance conditions through advertisement texts. The results of this study show that the Term Frequency- Inverse Document Frequency (TF-IDF) model combined with the Logistic Regression (LR) classifier obtained best results (F1 = .717), where 28% of the predictions were differentiating from the maintenance scores estimated by appraisers and other experts. Secondly, the real estate valuation model of Ortec Finance is used as a benchmark model to estimate the contribution of text mining methods to existing real estate valuation models. Compared to the benchmark model using the original maintenance scores, which were set manually by appraisers, the model using text mining methodologies did not improve significantly. However, it did show an increase of the accuracy by 5.26%. This study concludes that there is certain potential of applying text mining methods for real estate valuation in terms of automatization and explainability of real estate maintenance estimations. In addition, it contributes to normalize such practices and therefore shows potential for objectifying maintenance estimations. While considering the limitations and recommendations of this research, Ortec Finance could leverage this text mining classification tool to assist municipalities, appraisers and other experts for correcting real estate maintenance conditions in a more objective manner.