Machine Learning Techniques for House Valuations
Machine Learning Techniques for House Valuations, a project in which we investigate the use of machine learning on the house valuation problem.
The goal of this project is to accurately predict the price of a house, given its various attributes. The ‘black-box’ machine learning models like random forests and neural networks are typically quite accurate. However, accuracy is not the only consideration in house value estimation. Because of increasing regulations and demands from the customers, it is required to have a model that is economically interpretable, making it possible to explain how model values depend on house characteristics.
Interpretability & accuracy
The current econometric model of Ortec Finance strongly provides this: the model output can be decomposed into explainable temporal and hierarchical levels. With this dimension in mind, my research also investigates ‘hybrid’ methods in addition to black-box methods. The idea is to incorporate black-box machine learning functionalities to the econometric model of Ortec Finance in order to increase predictive accuracy while at the same time retaining as much interpretability as possible. In the end, we aim to have a thorough evaluation of traditional, hybrid and full black-box approaches to reveal the trade-off (or lack thereof) between interpretability and accuracy.
Ortec Finance has been offering solutions for house valuations for quite some time, and each year hundreds of thousands of objects are valued in the Netherlands, mostly for tax purposes. So, if we can improve on the current practice and method this would be highly beneficial for both the company and the customers.
Used Tools: R, Python and Ox