Comparative Performance of Machine Learning Classifiers in Detecting Performing and Non-performing Residential Property Renters

Abstract

In the recent decades, the use of artificial intelligence (AI) technology in decision making has continued to gain popularity in many disciplines including finance, marketing, insurance, engineering, and medicine to mention but a few, however, their applications have been very limited in the residential rental property market. The aim of the current paper is to compare the performance of four selected ML classifiers including Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and Logistic Regression (LR) with a view to choosing the best classifier appropriate for rental applications screening. A total of 724 data samples of the residential rental applications were obtained from the databases of 53 Estate Surveyors and Valuers (property managers), licensed to practice in the Lagos Metropolitan property market, Nigeria. The collected data were subdivided into training and testing datasets representing 70% and 30% respectively, and were analyzed using Python 3. The data were also used in determining the respective classification power of the classifiers using eight performance metrics such as recall/sensitivity, specificity, Type I error, Type II errors, and precision, others include F1 score, F1 adjusted measure, Mathew's Correlation Coefficient and area under the curve (AUC). The results reveal among others that the performance of all the four selected classifiers was good and satisfactory. However, DT outperformed the other classifiers in detecting true positive and false positive, while SVM achieved a better result than other classifiers in detecting false negative (Type II). As revealed in the study, the empirical comparison among different algorithms suggests that no single classifier is best for all learning tasks. The models provide cost and time-saving inputs for property investors, property management professionals, and policymakers.