Interpretable Machine Learning for the German residential rental market – shedding light into model mechanics

We compare the drivers in Machine learning models and give insights into their strengths and weaknesses predicting rental prices. The study employs SHAP values to measure feature importance. The study aims to investigate linear regression, decision tree and XGBoost algorithms. The research is unique in its application of IML methods to a large dataset of over 2.4 million observations in the German rental market and its application of comparative statistics using aggregate SHAP values. Main results are the superiority of XGB and LR showing higher SHAP values overall and thus explaining its lower predictive efficacy. DT models capture intricate interactions among variables with fewer features, while XGB accommodates more variables, emphasizing its higher complexity and thus superior performance. The top ten features for DT and XGB models show significant overlap, indicating robust concordance. Specific features are identified that distinguish the models, suggesting that a more complex model, like XGB, handles dummy variables more adeptly.