Loan Default Prediction · Credit Risk Dashboard

Total Loans

5,000

Synthetic portfolio

Default Rate

33.7%

1,686 defaults

Gini (GB)

0.491

Industry: >0.4 = good

KS Statistic

0.3809

Industry: >0.3 = good

ROC-AUC

0.7455

Gradient Boosting

LR Gini

0.521

Logistic Regression

5-Fold CV AUC

0.759

±0.009 std

High-Risk

759

PD score ≥ 65

Feature Importance · Gradient Boosting Top Drivers

Information Value (IV) Table · Predictive Power of Features

Feature	IV Score	Predictive Strength	Credit Risk Rationale

Model Performance Comparison

▶ GRADIENT BOOSTING (PRIMARY)

ROC-AUC0.7455

Gini Coefficient0.491

KS Statistic0.3809

PR-AUC0.5868

5-Fold CV AUC (mean ± std)0.7585 ± 0.0086

▶ LOGISTIC REGRESSION (SCORECARD)

ROC-AUC0.7607

Gini Coefficient0.5213

KS Statistic0.3948

Confusion Matrix · Gradient Boosting · Test (1,000 loans)

560

TRUE NEGATIVE

103

FALSE POSITIVE

182

FALSE NEGATIVE

155

TRUE POSITIVE

Note: Realistic AUC (0.745) and Gini (0.489) reflect production-grade model on noisy data — unlike synthetic fraud models (AUC 1.00). Gini >0.4 and KS >0.3 are industry-standard thresholds for a deployable credit scorecard.