Page 610 - AI for Good Innovate for Impact
P. 610
AI for Good Innovate for Impact
Use Case- 10: Dynamic Threshold Adjustment for Credit Scoring:
An Adaptive Optimization Approach for Subprime Lending
Organization: Vellore Institute of Technology, Chennai
Country: India
Contact Persons:
Parth Khairnar, pskhairnar1024@ gmail .com
Dr. Geetha S, geetha.s@vit.ac.in
1 Use Case Summary Table
Item Details
Category Finance
Traditional static credit thresholds fail to account for individual behavior
Problem Addressed
changes or economic shifts, resulting in unfair credit access denial.
• Adaptive thresholding using Markov Chain Monte Carlo (MCMC)
• Predictive modeling with Light Gradient Boosting Machine(GBM)
Key Aspects of Solu-
tion • Hyper-parameter tuning via Bayesian Optimization
• Receiver Operating Characteristic - Area Under the Curve(ROC-
AUC) metrics
LightGBM, MCMC, Bayesian Optimization, Hyperopt, Credit Risk Model-
Technology Keywords
ing, Fair AI, Python
Data Availability Public [1]
Metadata (Type of Text
Data)
• LightGBM model trained with K-Fold Cross Validation
Model Training and • MCMC for posterior sampling
Fine-Tuning
• Bayesian Optimization for hyper-parameter tuning
Testbeds or Pilot Initial testing in Google Colab with synthetic simulations
Deployments Potential for sandbox testing with financial partners [2]
2 Use Case Description
2�1 Description
Credit scoring has traditionally relied on logistic regression, decision trees, and more recent
methods like random forests or neural networks to calculate an explicit probability threshold
574
