Credit Scoring and Default Prediction

Module: Fintech AI | Difficulty: Advanced

Logistic Regression

Credit Scorecard

Survival Model for Default

Evaluation

| Metric | Interpretation | |--------|---------------| | AUC-ROC | Discrimination | | Gini | | | KS | Maximum separation | | PSI | Population stability |

import numpy as np
from sklearn.linear_model import LogisticRegression

class CreditScorer:
    def __init__(self):
        self.model = LogisticRegression(penalty='l1', C=0.1)
    def fit(self, X, y):
        self.model.fit(X, y)
    def predict_proba(self, X):
        return self.model.predict_proba(X)[:, 1]
    def score_to_odds(self, score, factor=20, offset=500):
        odds = np.exp((score - offset) / factor)
        return odds / (1 + odds)
    def gini(self, y_true, y_pred):
        from sklearn.metrics import roc_auc_score
        return 2 * roc_auc_score(y_true, y_pred) - 1
    def ks_statistic(self, y_true, y_pred):
        from scipy.stats import ks_2samp
        pos = y_pred[y_true == 1]
        neg = y_pred[y_true == 0]
        return ks_2samp(pos, neg).statistic

Research Insight: Credit scoring requires explainability — regulations (ECOA, FCRA) require that decisions be explainable. While ML models achieve higher AUC, logistic regression remains dominant because of its interpretability. SHAP values help explain complex models.