1. Introduction

Banks and fintechs must monitor not only fraudulent payments but also customers whose behaviour drifts from historical norms—such as large night transfers on new devices, sudden destination sprawl, declining cashflow, or elevated ATO signals. Transaction fraud models answer “is this event fraud?”; credit risk models answer “is this customer becoming high risk?” Both layers complement each other in an enterprise AI risk platform.

This document presents the Nexius Credit Scoring module: a supervised customer-scoring pipeline from master-data feature aggregation through training, evaluation, grading, and serving. The contribution is a transparent, bank-oriented research protocol that makes imbalance-aware and operational choices explicit.

2. Problem Formulation

Let each customer $c$ be represented by an aggregated feature vector $x_c$. We estimate $s_c = P(\text{high_risk} \mid x_c) \in [0, 1]$. A policy $\pi$ maps scores to grades {APPROVE, REVIEW, DECLINE} (default cutoffs approximately 0.4 and 0.7). Research evaluation emphasizes ranking metrics on a stratified held-out set; operational evaluation reports confusion-based metrics at the primary high-risk threshold and early-warning rule coverage.

3. Data and Feature Engineering

3.1 Data Sources

Features are aggregated from master tables:

• customers
• accounts
• transactions
• logins
• devices
• SLIK
• fraud_labels
• statement_lines

Institutions can substitute equivalent warehouse extracts.

3.2 Feature Groups

• KYC profile: PEP, monthly income, labeled risk, SLIKed.
• Accounts & cashflow: account count, dormant accounts, average balance, cashflow slope from statement lines.
• Transaction behaviour: counts and amount statistics, night ratio, unique destinations, transfer ratio, spend-versus-income.
• Access & fraud history: login volume, failed logins, device count, rooted devices, prior fraud transaction count.

3.3 Label Construction

Positive label is_default is true when KYC risk level is HIGH, the customer is SLIKed, or severe delinquency/repayment stress triggers (elevated night activity, many unique destinations, ATO-like login failures, rooted devices, prior fraud transactions, sharply declining cashflow). Remaining customers are negatives. Heuristic positives are documented as research proxies—not substitutes for investigator-confirmed risk ratings in production studies.

4. Learning Algorithm

A binary classifier is trained with LightGBM when available; otherwise, HistGradientBoosting is used. Class imbalance is handled via class_weight=balanced. Tree ensembles suit tabular banking aggregates, support fast research iteration, and enable permutation-importance attribution. The currently deployed artifact reports model type: HistGradientBoostingClassifier.

5. Experimental Protocol

5.1 Stratified Split

Stratified holdout with configurable test_size is used to preserve high-risk prevalence across train/test splits. When class counts allow, stratified K-fold PR-AUC provides additional stability check. Split metadata is reported explicitly (active run: time, n_train=800, n_test=200).

5.2 Metrics

Accuracy is excluded as a headline KPI under imbalance. The focus is on PR-AUC and ROC-AUC. At a threshold tuned toward F1 with a precision preference, we report Precision, Recall, and F1 for the high-risk class.

Latest holdout snapshot in this deployment: PR-AUC=1.000, ROC-AUC=1.000.

5.3 Bucket Policy

Continuous scores map to:

• APPROVE: PD < 0.2
• REVIEW: PD < 0.45
• DECLINE: PD ≥ high cutoff (default near 0.4 / 0.7)

Separating scoring from grades lets research study model quality independently from risk appetite.

5.4 Early Warning Overlays

Rule-based early warnings (declining cashflow, high night ratio, many unique destinations, repeated failed logins, prior fraud history) accompany scores so analysts receive actionable narratives beyond a raw probability.

6. Research Use Cases

• Imbalance-aware benchmarking: Compare PR-AUC vs. ROC-AUC as high-risk prevalence changes.
• Feature-family ablation: Remove KYC, velocity, or access groups to quantify lift for investigation hypotheses.
• Bucket stress tests: Sweep medium/high cutoffs against review-team capacity.
• Label sensitivity: Measure degradation when heuristic positives replace investigator ratings.
• Cohort drift: Retrain on rolling windows to study channel-mix and behaviour-shift effects.

7. Deployment Notes

Artifacts persist as model bundles with metrics metadata. Batch scoring produces scored customers for dashboards; online serving exposes GET /ai/credit/scores and overview aggregates. Dashboards should show full-population aggregates computed server-side; detail tables remain paginated. Human-in-the-loop review is always required for DECLINE decisions in regulated settings.

8. Limitations

Results on synthetic master data are not external validity claims for any bank. Heuristic labels, delayed KYC updates, and incomplete device or statement coverage can inflate research metrics. Accuracy is deliberately omitted as a headline KPI under imbalance.

9. Conclusion

We formalised a customer-level credit risk research pipeline aligned with banking operations: aggregated tabular features, imbalance-aware boosting, stratified evaluation, PR-centric metrics, an explicit APPROVE/REVIEW/DECLINE grade policy, and early-warning overlays. The protocol supports reproducible experimentation and incremental productionisation within an enterprise AI risk platform.

References

• Davis, J., & Goadrich, M. (2006). The relationship between Precision-Recall and ROC curves. ICML.
• Ke, G., et al. (2017). LightGBM: A highly efficient gradient boosting decision tree. NeurIPS.
• Bolton, R. J., & Hand, D. J. (2002). Statistical fraud detection: A review. Statistical Science.
• Chandola, V., Banerjee, A., & Kumar, V. (2009). Anomaly detection: A survey. ACM Computing Surveys.
• Nexius AI Lab. (2026). Enterprise AI Risk Platform — Credit Scoring Module Technical Note (internal).