Abstract
This article addresses the problem of classifying tenant reliability using machine learning methods. Leveraging a rich dataset with diverse features such as average debt, total number of registered complaints, rental duration in months, number of successfully resolved complaints, and other parameters, a classification task was formulated. A comparative analysis of classification models—BERT, Random Forest, and K-Nearest Neighbors—was conducted to assess tenant reliability. The transformer-based model (BERT), after fine-tuning, demonstrated the highest performance across metrics: ROC-AUC (0.95), Precision (0.94), Recall (0.96), and F1-score (0.95). The results indicate the model's high applicability for rental risk assessment tasks. The proposed approach highlights the potential of advanced transformer-based models in addressing real-world social and economic challenges.