A Comparative Study of Traditional Credit Scoring Models and Predictive Analytics Models in Reducing Non-Performing Loans
  • Author(s): Ehimare Ucheoma Austie
  • Paper ID: 1710822
  • Page: 530-536
  • Published Date: 28-02-2024
  • Published In: Iconic Research And Engineering Journals
  • Publisher: IRE Journals
  • e-ISSN: 2456-8880
  • Volume/Issue: Volume 7 Issue 8 February-2024
Abstract

All over the world, financial institutions continue to suffer profound losses due to the ubiquitous threat posed by non-performing loans (NPLs). Recent research suggests that NPLs persist at an alarming rate of 3-5% across both emerging and developed economies, especially after the global crisis wrought by the COVID-19 pandemic and the consequent inflationary onslaught. This research is a historic comprehensive analysis that benchmarks and compares the traditional credit scoring methodologies, including but not limited to, logistic regression, linear discriminant analysis, and various scorecard systems, against emerging cognitive predictive analytics and Machine Learning (ML) disciplines, such as random forests and gradient boosting machines (like XGBoost) and neural networks. This research leverages existing regulation, such as Basel III, together with the IMF’s Global Financial Stability Reports, extensive primary research, and industry research from the Bank for International Settlements (BIS) as well as the growing body of systematic review literature on ML in credit risk management to accomplish several objectives: Through quantitative benchmarks such as area- under- the- curve receiver operating characteristic (AUC-ROC), predictive models are quantitatively distinguished from traditional models in terms of architecture and operability, as well as performance, in the predictive analytic models and their traditional counterparts, which are AUC-ROC metrics for predictive default classification. These traditional models are sullied by missing and null values, as well as echoing warnings of rising and falling national debts. In the proposed multifaceted organizational readiness assessment of the hybrid model coupled with deep learning, phasic granular strategies for model integration are emphasized to maximize the synergies of the integrated models. The organization’s capability to synthesize the knowledge of 120 credit risk professionals and three in-depth comparative readiness assessment models is from the survey. In it, the quantifiable benefits of NPL reduction and the multifaceted barriers to implementation are outlined succinctly within the synthesized framework. In tandem with the proposed strategy, it is empirically validated the traditional models are 15-25% less predictive in key metrics such as NPL ratio through AUC-ROC predictive default classification of 10-20% ratio reduced simulated portfolios. The hybrid and deep learning models increase EWS and risk dynamic stratification. Though there is granular evidence confirming the enhanced model’s performance, issues still remain. The model’s data quality issues, the black box nature of Machine Learning models, the interdisciplinary adeptness within the bounded compliance of model legacies to evolving standards of AI governance, the ever-saturated compliance with the model-orange complex under the deeply rooted structural constraints of model deployment and maintenance. Ultimately, this research posits that with meticulous governance structures, including bias audits and explainable AI (XAI) integrations, hybrid approaches combining the interpretability of traditional models with the predictive power of analytics can revolutionize credit risk management, fostering more resilient lending ecosystems, optimizing capital allocation, and bolstering overall financial stability in an increasingly volatile economic landscape.

Keywords

Traditional Credit Scoring, Predictive Analytics, Machine Learning Ensembles, Non-Performing Loans, Credit Risk Management, Hybrid Model Integration

Citations

IRE Journals:
Ehimare Ucheoma Austie "A Comparative Study of Traditional Credit Scoring Models and Predictive Analytics Models in Reducing Non-Performing Loans" Iconic Research And Engineering Journals Volume 7 Issue 8 2024 Page 530-536

IEEE:
Ehimare Ucheoma Austie "A Comparative Study of Traditional Credit Scoring Models and Predictive Analytics Models in Reducing Non-Performing Loans" Iconic Research And Engineering Journals, 7(8)