Comparative Analysis of Machine Learning Methods for Predicting Property Prices and Sale Velocities in the Real Estate Industry

Abstract: The real estate industry is one of the largest industries in the world and using data-driven decision-making has been shown to increase companies’ profitability. A technique to apply data-driven decision-making is machine learning. Within the real estate industry, predicting property selling prices and sale velocities (the duration a property remains on the market) are crucial factors of interest. Knowing the selling price and the sale velocity can motivate businesses to alter their plans in an effort to increase their profitability. The research conducted in this thesis employs a comparative approach to evaluate the performance of various machine learning methods in predicting both the selling price and the sale velocity of properties. The machine learning methods this study investigated are random forest, decision tree, K-nearest neighbor, support vector regression, and multilayer perceptron. After pre-processing, the data set used comprises 560,000 distinct data points from the Swedish housing market. The data set has a wide geographic scope, covering almost the entire country of Sweden. The data set was subjected to both normalization and standardization techniques in order to determine how they affected the machine learning methods. The results demonstrate that random forest oEutperforms the other machine learning methods in predicting property selling prices. However, the assessed machine learning methods encountered difficulties in predicting the sale velocity. The best-performing machine learning method for sale velocity is random forest. Notably, SVR demonstrates a lower MAE for sale velocity, but performs worse in the R² metric.