Predicting Default Probability in Emerging Economies: A Comparative Machine Learning Approach in the Tunisian Banking Sector

Aug 4, 2026

Upload by Violet

Economic policymakers face serious risks when default exposure is ignored, undermining financial system stability and wider macroeconomic resilience. This dissertation applies advanced machine learning to predict loan default probabilities for Tunisian banking entities listed in the Tunisian central bank dataset from 2018 to 2022. Two modeling strategies are compared: continuous-variable-based features and ratio-focused features. Results show strong performance from both approaches, with Random Forest combined with SVM and SMOTE or Tomek Links resampling identified as the most effective. The study supports credit risk assessment in emerging markets.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages53

Reading Time2 h 14 min

AccessFree to download

Introduction

Literature review

Methodology

Data description and understanding

Data preparation

Model training and evaluation

Results and discussion

Conclusion and future work

Master Degree Program in Data Science and Advanced Analytics PREDICTING DEFAULT PROBABILITY IN EMERGING ECONOMIES: A COMPARATIVE MACHINE LEARNING APPROACH IN THE TUNISIAN BANKING SECTOR Mohamed Ali FELFEL Dissertation presented as partial requirement for obtaining the Master Degree Program in Data Science and Advanced Analytics NOVA Information Management School Instituto Superior de Estatística e Gestão de Informação Universidade Nova de Lisboa NOVA Information Management School Instituto Superior de Estatística e Gestão de Informação Universidade Nova de Lisboa PREDICTING DEFAULT PROBABILITY IN EMERGING ECONOMIES: A COMPARATIVE MACHINE LEARNING APPROACH IN THE TUNISIAN BANKING SECTOR by Mohamed Ali Felfel Dissertation presented as partial requirement for obtaining the Master’s degree in Advanced Analytics, with a Specialization in Data Science Supervisor: Afshin Ashofteh February 2024 STATEMENT OF INTEGRITY I hereby declare having conducted this academic work with integrity. I confirm that I have not used plagiarism or any form of undue use of information or falsification of results along the process leading to its elaboration. I further declare that I have fully acknowledge the Rules of Conduct and Code of Honor from the NOVA Information Management School. Mohamed Ali Felfel Lisbon, 29/02/2024 ACKNOWLEDGEMENTS To my academic mentor, Prof. Afshin Ashofteh for providing me the opportunity to engage in this project. I extend my gratitude for his continuous guidance, encouragement, and unwavering commitment throughout the duration of this thesis. Furthermore, I am indebted to my parents whose unwavering efforts have made this journey possible. I cannot thank them enough for their constant belief in me, unconditional love, and sacrifices. A special mention goes to my cherished brother for his consistent encouragement and positive energy. In addition, my heartfelt thanks go out to my entire family and friends in Tunisia, Portugal, and across the globe. I appreciate the moral support and constant inspiration they provided. Their unwavering belief in me has been a driving force, keeping my spirits high and motivation intact during this process. The shared beautiful moments hold significant value. This work is dedicated to everyone who played a role in making this incredible experience happen, contributing to shaping the person I am today. Having you all by my side fills me with gratitude, and I sincerely thank you for your unwavering support, love, and care. ABSTRACT Economic policymakers neglecting default risk can have severe consequences for the financial system and broader economic stability. This research addresses this critical challenge by deploying advanced machine learning models for loan default prediction, aiming to predict the default probabilities of companies listed in the Tunisian central bank between 2018 and 2022. Two distinct approaches were implemented to enhance the predictive accuracy of default probabilities. The first approach utilized a comprehensive set of continuous variables, capturing various financial indicators. The second approach focused on key financial ratios, providing a more nuanced understanding of companies' financial health. By employing these approaches, this research not only aimed to calculate default probabilities but also sought to compare the predictive power of both methodologies. The results of this study reveal promising findings, indicating excellent performance by both approaches in predicting default rates. Furthermore, the research identified Random Forest models with SVM SMOTE and Tomek Links resampling techniques as the most effective in each respective approach. These findings underscore the importance of leveraging advanced machine learning techniques for credit risk assessment, particularly in emerging economies where data environments may present challenges. This work contributes significantly to the field by demons

FAQs

What problem does the dissertation address?

It tackles the consequences of neglecting default risk by predicting loan default probabilities using advanced machine learning for Tunisian banking sector entities.

Which time period and dataset are used?

The study predicts default probabilities for companies in the Tunisian central bank between 2018 and 2022.

How are the two approaches different?

One approach uses a comprehensive set of continuous variables to capture financial indicators, while the other focuses on key financial ratios to reflect companies' financial health.

What models and techniques performed best?

Random Forest models using SVM SMOTE and Tomek Links resampling techniques showed the strongest effectiveness within each respective approach, achieving excellent prediction of default rates.

Aug 4, 2026

Upload by Violet