Learning Objectives:

  • Understand the different types of credit scoring models used in digital lending.

  • Explain the role of AI and machine learning in credit scoring.

  • Understand the use of alternative data in credit scoring.

  • Analyse model interpretability and responsible AI in credit decisions.

4.1 Types of Credit Scoring Models

Credit scoring models assess the creditworthiness of borrowers. Traditional models have relied on credit bureau data. AI-powered models leverage advanced machine learning techniques and alternative data sources.

Traditional Models:

  • Logistic Regression: A traditional statistical model used for credit scoring.

  • Credit Bureau Scores: Scores generated by credit bureaus, such as FICO or CIBIL. The IIBF CAIIB ABM curriculum identifies “CIBIL Score” as a key parameter in the Credit Underwriting Engine .

AI/ML Models:

  • Decision Trees: A machine learning model that splits data based on decision rules.

  • Random Forests: An ensemble of decision trees that improves prediction accuracy. The NobleProg course covers “Random forests” for credit scoring .

  • Gradient Boosting: A powerful machine learning technique that builds models sequentially. The NobleProg course covers “Gradient boosting (LightGBM, XGBoost) for scoring accuracy” .

  • Logistic Regression: The HKSI course identifies logistic regression as a traditional ML technique lenders use to assess credit risk .

The Role of AI in Credit Scoring:
The HKSI course covers “AI & ML Techniques in Retail Credit Risk” and “Credit Origination & Underwriting,” describing how AI is being leveraged throughout the entire retail credit lifecycle, from rapid approvals to dynamic portfolio monitoring . The NobleProg course covers “Machine Learning for Credit Scoring” including logistic regression, decision trees, random forests, and gradient boosting .

4.2 Alternative Data in Credit Scoring

The use of alternative data is a key innovation in digital lending. The MIT FinTech course covers “Alternative Data Across the Loan Life Cycle” as a core topic, examining how FinTechs use alternative data to assess creditworthiness . The UIBFS syllabus includes “Emerging Trends in Credit and Lending” covering “FinTech and digital lending, Peer-to-peer lending and crowdfunding” .

Types of Alternative Data:

  • Transactional Data: Bank account transaction history, spending patterns.

  • Behavioral Data: Mobile-usage patterns, social media data, online behaviour.

  • Utility Payments: Payment history for utilities, rent, and other bills.

  • E-Commerce Data: Purchase history and transaction data from e-commerce platforms.

  • Employment Data: Employment history and stability.

How Alternative Data is Used:

  • Credit Scoring: Improving credit scores for applicants without formal credit histories.

  • Risk Assessment: Enhancing risk prediction through additional data points.

  • Fraud Detection: Identifying anomalies and potential fraud.

The NobleProg course covers “Data for Credit Scoring Models,” including “Sources: transactional, behavioral, and alternative data,” “Data cleaning and feature engineering for lending decisions,” and “Handling class imbalance and data scarcity in risk prediction” .

4.3 Model Interpretability and Responsible AI

Model interpretability is a key consideration in AI-powered credit underwriting. The HKSI course identifies the “key benefits and limitations/challenges associated with using AI for credit risk assessment” . The NobleProg course covers “Model Interpretability and Responsible AI” including “Explaining predictions with SHAP and LIME,” “Fairness in credit models: bias detection and mitigation,” and “Compliance with regulatory frameworks (e.g. ECOA, GDPR)” .

Key Challenges:

  • Bias: The HKSI course covers “limitations/challenges associated with using AI for credit risk assessment” . AI models may perpetuate or amplify existing biases.

  • Explainability: The NobleProg course covers “Explaining predictions with SHAP and LIME” . “Black box” models can be difficult to interpret for regulatory compliance.

  • Regulatory Compliance: The NobleProg course covers “Compliance with regulatory frameworks (e.g. ECOA, GDPR)” . AI decisions must comply with fair lending laws and data protection regulations.

PreviousNext