Scholarlink

Financial institutions face persistent challenges in identifying potential defaults and enhancing recovery strategies. This study develops and evaluates a data-driven machine learning pipeline leveraging classification models and clustering-based borrower segmentation. A real-world dataset comprising 500 Indian loan records with borrower demographic, transactional, and behavioral features is analyzed. Multiple classifiers, including Logistic Regression, Support Vector Machines (SVM), Decision Tree, and Random Forest, are trained to predict recovery risk. Complementary unsupervised clustering provides borrower segmentation based on loan amount and income. The Logistic Regression model achieves perfect classification with 100% precision, recall, and AUC-ROC, outperforming all other models. Comparisons with recent literature show our approach exceeds prior benchmarks in both interpretability and predictive power. The model is deployable, interpretable, and highly accurate, offering significant operational value for financial recovery management.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Robust Predictive Modelling for Loan Recovery Risk Assessment Using Supervised Learning and Clustering Techniques

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sanskruti Vadakattu,
  • Dhruvi Dombe,
  • Yash Vardhan Agrawal,
  • Ashok Sarabu,
  • T. Aditya Sai Srinivas

摘要

Financial institutions face persistent challenges in identifying potential defaults and enhancing recovery strategies. This study develops and evaluates a data-driven machine learning pipeline leveraging classification models and clustering-based borrower segmentation. A real-world dataset comprising 500 Indian loan records with borrower demographic, transactional, and behavioral features is analyzed. Multiple classifiers, including Logistic Regression, Support Vector Machines (SVM), Decision Tree, and Random Forest, are trained to predict recovery risk. Complementary unsupervised clustering provides borrower segmentation based on loan amount and income. The Logistic Regression model achieves perfect classification with 100% precision, recall, and AUC-ROC, outperforming all other models. Comparisons with recent literature show our approach exceeds prior benchmarks in both interpretability and predictive power. The model is deployable, interpretable, and highly accurate, offering significant operational value for financial recovery management.