Calibrated Credit Intelligence: Shift-Robust and Fair Risk Scoring with Bayesian Uncertainty and Gradient Boosting

Paper: arXiv 2603.06733

Authors: Srikumar Nayak

Abstract

Credit risk scoring must support high-stakes lending decisions where data distributions change over time, probability estimates must be reliable, and group-level fairness is required. While modern machine learning models improve default prediction accuracy, they often produce poorly calibrated scores under distribution shift and may create unfair outcomes when trained without explicit constraints. This paper proposes Calibrated Credit Intelligence (CCI), a deployment-oriented framework that combines (i) a Bayesian neural risk scorer to capture epistemic uncertainty and reduce overconfident errors, (ii) a fairnessconstrained gradient boosting model to control group disparities while preserving strong tabular performance, and (iii) a shiftaware fusion strategy followed by post-hoc probability calibration to stabilize decision thresholds in later time periods. We evaluate CCI on the Home Credit Credit Risk Model Stability benchmark using a time-consistent split to reflect real-world drift. Compared with strong baselines (LightGBM, XGBoost, CatBoost, TabNet, and a standalone Bayesian neural model), CCI achieves the best overall trade-off between discrimination, calibration, stability, and fairness. In particular, CCI reaches an AUC-ROC of 0.912 and an AUC-PR of 0.438, improves operational performance with Recall@1%FPR = 0.509, and reduces calibration error (Brier score 0.087, ECE 0.015). Under temporal shift, CCI shows a smaller AUC-PR drop from early to late periods (0.017), and it lowers group disparities (demographic parity gap 0.046, equal opportunity gap 0.037) compared to unconstrained boosting. These results indicate that CCI produces risk scores that are accurate, reliable, and more equitable under realistic deployment conditions.

Complexity vs Empirical Score

• Math Complexity: 6.5/10
• Empirical Rigor: 8.0/10
• Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper provides a well-structured framework combining Bayesian neural networks and fairness-constrained gradient boosting to tackle real-world credit risk challenges like distribution shift and calibration. It features strong empirical methodology using a public benchmark dataset with time-consistent temporal splits and rigorous probability calibration.

Research Flowchart

  flowchart TD
    A["Research Goal: Robust, Fair, and Calibrated Credit Risk Scoring"] --> B["Data: Home Credit Benchmark (Time-Consistent Split)"]
    
    subgraph C ["Calibrated Credit Intelligence (CCI) Framework"]
        C1["Bayesian Neural Network (Epistemic Uncertainty)"]
        C2["Fairness-Constrained Gradient Boosting (Tabular Performance)"]
        C1 & C2 --> C3["Shift-Aware Fusion & Post-Hoc Calibration"]
    end
    
    B --> C
    C --> D["Key Outcomes"]
    
    subgraph D ["Key Outcomes"]
        D1["High Accuracy: AUC-ROC 0.912, Recall@1%FPR 0.509"]
        D2["Low Calibration Error: Brier 0.087, ECE 0.015"]
        D3["Robust to Drift: AUC-PR Drop 0.017"]
        D4["Improved Fairness: DP Gap 0.046, EO Gap 0.037"]
    end

Cited References (15)

1. Credit risk modeling using transfer learning and domain adaptation (2022)
2. Machine learning for credit scoring: Improving logistic regression with non-linear decision-tree effects (2022)
3. A two-stage hybrid credit risk prediction model based on XGBoost and graph-based deep neural network (2022)
4. Hybrid boosted attention-based LightGBM framework for enhanced credit risk assessment in digital finance (2025)
5. Enhancing financial risk prediction for listed companies: A CatBoost-based ensemble learning approach (2024)
6. Advanced user credit risk prediction model using lightgbm, xgboost and tabnet with smoteenn (2024)
7. Investment decision making for large-scale Peer-to-Peer lending data: A Bayesian Neural Network approach (2025)
8. Toward Fairness-Aware Gradient Boosting Decision Trees for Ranking

0 of 15 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.