Explainable AI in Loan Decision Modeling

This analysis explores a loan denial decision through the lens of explainable AI (XAI), presented in a courtroom-style format. The case centers on Jane Dow, a 37-year-old professional woman with a bachelor’s degree and an executive occupation, whose loan application was denied by a machine learning model.

📖 Project Overview

• Objective: Investigate why the model denied Jane’s loan and analyze whether the reasoning was consistent and fair.
• Approach: Apply three leading XAI techniques (SHAP, LIME, Anchors) to interpret the decision.
• Format: Courtroom-style analysis with Prosecution (argues unfairness) and Defense (justifies the denial).
• Assigned Role: I was assigned the Defense, focusing on consistency and reliability of the model’s reasoning.

🛠️ Methods and Tools

• Dataset: UCI Adult Income dataset
• Model: Random Forest Classifier (scikit-learn)
• XAI Techniques:
  • SHAP (SHapley Additive exPlanations)
  • LIME (Local Interpretable Model-agnostic Explanations)
  • Anchors (High-Precision Model-Agnostic Explanations)

📊 Results

• SHAP: Showed both positives (education, hours worked, occupation) and negatives (marital/relationship, no capital gains).
• LIME: Capital gain absence was the strongest negative; probability of approval is 30%.
• Anchors: Denial rule applied consistently with 95+% precision.