// ACADEMIC CONTENT PORTAL · Computer Science

Thesis Topic Ideas: Explainable AI for Loan-Default Prediction in Nepali Microfinance

Level: PhDDifficulty: Advanced

Applicable Universities:Kathmandu University (KU)Tribhuvan University (TU)Pokhara University (PU)University of OxfordUniversity of Cambridge

🎓 Academic Journey Progression Map

Grade 8-10 (SEE)
School Projects
+2 (Grades 11-12)
Stream Projects
Bachelor's Degree
Undergraduate Thesis
Master's Degree
Postgraduate Thesis
PhD Doctorate
Dissertation Guidance

1. Introduction & Problem Statement

Overview: PhD-level research developing interpretable machine learning credit-risk models (SHAP/LIME) on uncollateralized microfinance data, adhering to Nepal Rastra Bank transparency mandates.

Background Context (Nepal): Black-box ML credit scoring models risk perpetuating algorithmic bias against marginalized rural borrowers. Implementing Explainable AI (XAI) provides feature importance transparency required by NRB regulatory guidelines.

2. Research Objectives

›Curate and anonymize longitudinal credit dataset from Class 'D' Microfinance Financial Institutions (MFIs) in Nepal
›Develop ensemble prediction models (XGBoost, LightGBM, Neural Networks) for uncollateralized loan default risk
›Apply model-agnostic Explainable AI frameworks (SHAP, LIME) to generate individual loan decision explanations
›Evaluate algorithmic fairness metrics across gender, geographic region, and ethnic borrower demographics
›Formulate regulatory compliance framework for transparent AI credit scoring under NRB oversight

3. Proposed Methodology

Data collection and preprocessing partnering with MFI networks under strict non-disclosure ethics agreements
Supervised machine learning training with SMOTE oversampling for imbalanced default class labels
Local and global interpretability analysis using Shapley Additive exPlanations (SHAP) and LIME
Demographic parity and equalized odds fairness auditing across rural female borrower cohorts
Rigorous model validation evaluating ROC-AUC, Precision-Recall AUC, and interpretability fidelity

$ Worked Example / Sample Scenario

Sample Scenario: For an MFI applicant flagged as 'High Default Risk', the SHAP waterfall plot reveals that seasonal harvest cash-flow volatility contributed +0.34 to risk, while 100% peer-group repayment history offset risk by -0.18, giving credit officers clear transparent rationale.

4. Thesis Chapter-by-Chapter Outline

Chapter 1: Introduction & Research GapTU/KU/NEB standard

Algorithmic credit scoring in microfinance, explainability imperatives, and PhD research goals

Chapter 2: Literature & Theoretical FrameworkTU/KU/NEB standard

Credit risk literature, interpretable AI algorithms (SHAP/LIME), and NRB regulatory landscape

Chapter 3: Methodology & XAI ArchitectureTU/KU/NEB standard

Data preprocessing, ML ensemble training, fairness metrics, and SHAP feature attribution setup

Chapter 4: Empirical Results & InterpretabilityTU/KU/NEB standard

Default prediction accuracy, SHAP waterfall plots for sample borrowers, and fairness audit outcomes

Chapter 5: Conclusion & Regulatory FrameworkTU/KU/NEB standard

PhD theoretical contributions, policy recommendations for NRB, and future research directions

5. Recommended Tools & Technologies

To implement the practical, technical, or analytical portions of this thesis topic, the following software tools, libraries, or APIs are recommended:

PythonPyTorchSHAPLIMEXGBoostScikit-Learn

6. Core References & Academic Sources

[1]Nepal Rastra Bank — Unified Directives for Microfinance Financial Institutions (Class 'D')
[2]IEEE Transactions on Knowledge and Data Engineering — Explainable AI in Financial Risk
[3]Journal of Banking & Finance — Machine Learning Credit Scoring in Emerging Markets

7. Frequently Asked Questions (FAQs)

Q: What is the difference between SHAP and LIME in XAI?

SHAP calculates Shapley values from game theory for global and local consistency, while LIME builds local linear surrogate models around individual predictions.

Q: Why is explainability essential for microfinance credit scoring?

Central bank regulations mandate transparent non-discriminatory lending rationale, preventing black-box rejection of rural borrowers.

Q: How do you handle severe class imbalance in loan default datasets?

Techniques like SMOTE, Focal Loss, and Precision-Recall AUC optimization handle sparse default occurrences.

Q: Can SHAP explanations be deployed in real-time MFI mobile apps?

TreeSHAP computes explanations in milliseconds for tree models, enabling instant explanation generation on server APIs.