Comparison of different machine learning methods’ capability to predict housing prices

Sammanfattning: Accurate evaluations in the real estate market are valuable for many different parties, including lenders, agents, and buyers. Achieving an accurate evaluation today is challenging, even with good knowledge of the market. This report is analysing a machine learning approach to evaluation in the Stockholm city housing market and includes a comparison between Random Forest-, K Nearest Neighbour Regression and a Neural Network implementation. The performance for each model is measured in terms of the difference between the actual sold price and the predicted price as well as the relative percentage error and R² value. The Random Forest implementation was the one with the best performance with an absolute error of 354650 SEK, a percentage error of 7.60 %, and a R² value of 0.951. The result yielded by the models in this study demonstrates the potential of machine learning-based appraisals. With further improvements in data and algorithms, most parties can start relying on automated appraisal models instead of traditional property valuation.