Credit Risk Scoring & Explainable AI Intelligence Platform

Executive Summary

The Credit Risk Scoring & Explainable AI Platform is an enterprise-grade, quantitative underwriting and credit intelligence solution. It bridges rigorous Basel II/III internal ratings-based (IRB) credit risk modeling with Explainable AI (TreeSHAP / DiCE Counterfactuals) and modern, reactive full-stack web engineering.

Traditional "black-box" machine learning models often deliver superior discriminatory power ($AUC \approx 0.78+$) compared to legacy scorecards ($AUC \approx 0.73$), but fail regulatory compliance due to opacity. This platform addresses this trade-off by combining:

High-Discrimination GBDT Modeling: 5-Fold Stratified LightGBM achieving AUC-ROC = 0.7854 and KS-Statistic = 43.17%.
Standard FICO Point Calibration: Mathematical log-odds mapping onto a classic 300–850 credit score range ($\text{Base}=600$, $\text{PDO}=20$).
Adverse Action Reason Codes (FCRA / ECOA): Automated extraction of top negative risk drivers mapped into legally compliant adverse notices.
Actionable Counterfactual Recourse (DiCE): Prescriptive financial adjustments (e.g., loan reduction, tenure extension) to flip rejections into approvals.
Macro Stress Testing & Data Drift (ICAAP / Basel III): Real-time Population Stability Index ($\text{PSI}$) tracking and economic scenario simulation to assess required Capital Buffer Deltas.
Production Microservices & UI: Docker-ready FastAPI backend with an immutable SQLite audit trail paired with an ultra-clean, bilingual (TH/EN) Next.js frontend.

End-to-End System Architecture

flowchart TD
    subgraph Data_Pipeline["1. Quantitative Data Pipeline"]
        RawData["Raw Data Ingestion<br/>(Home Credit / Bureau Parquet)"] --> Preprocessing["Data Cleaning & Outlier Capping<br/>(IQR 1.5x & Robust Imputation)"]
        Preprocessing --> FeatureEng["Domain Feature Engineering<br/>(DTI, Annuity Ratio, Volatility)"]
        FeatureEng --> WoE_IV["WoE Binning & IV Selection<br/>(VIF Multicollinearity Pruning)"]
    end

    subgraph Modeling_Core["2. Champion ML & Calibration Engine"]
        WoE_IV --> LGBM["Champion LightGBM<br/>(5-Fold Stratified Ensemble, Optuna)"]
        LGBM --> Calib["Isotonic Calibration & Log-Odds<br/>PD Estimation"]
        Calib --> FICO["FICO Scorecard Transform<br/>(300 - 850 Points, PDO=20)"]
    end

    subgraph XAI_Governance["3. XAI & Regulatory Governance"]
        LGBM --> SHAP["TreeSHAP Attribution Engine"]
        SHAP --> AdverseAction["FCRA / ECOA Adverse Action<br/>Reason Codes (RC_01 - RC_07)"]
        LGBM --> DiCE["DiCE Counterfactual Engine<br/>Actionable Recourse Optimization"]
        Calib --> PSI_Engine["Population Stability Index (PSI)<br/>& Macro Stress Testing (ICAAP)"]
    end

    subgraph Backend_Services["4. FastAPI Microservice Core"]
        FICO --> API_Score["POST /api/v1/score"]
        AdverseAction --> API_Score
        DiCE --> API_Score
        PSI_Engine --> API_Sim["POST /api/v1/simulate/policy<br/>GET /api/v1/simulate/stress-test"]
        API_Score --> SQLite_DB[("SQLite WAL Audit Trail<br/>(credit_risk_platform.db)")]
    end

    subgraph Frontend_App["5. Next.js 16 Client (1440px Widescreen)"]
        API_Sim --> Tab1["Tab 1: 🎯 Policy & Profit Simulator"]
        API_Score --> Tab2["Tab 2: 👤 Underwriting & Recourse Engine"]
        API_Sim --> Tab3["Tab 3: ⚡ Macro Stress & PSI Tracker"]
        SQLite_DB --> Tab2
    end

Quantitative Risk Modeling & Machine Learning Core

1. Data Engineering & Feature Pipeline

Missing Value Imputation: Domain-aware median imputation for skewed numerical attributes; missing categorical states tracked via explicit "Missing" level.
Outlier Treatment: Soft Winsorization capping extreme values at the 99th percentile for heavy-tailed income and transaction metrics.
Domain Financial Ratios:

$$\text{Credit Term} = \frac{\text{Total Credit Amount}}{\text{Monthly Installment}}$$

$$\text{Annuity-to-Income Ratio} = \frac{\text{Monthly Installment}}{\text{Total Income}}$$

$$\text{External Sources Mean} = \frac{\text{External Source 1} + \text{External Source 2} + \text{External Source 3}}{3}$$

Engineered Feature Definitions:

CREDIT_TERM = AMT_CREDIT / AMT_ANNUITY (Loan duration proxy)
ANNUITY_INCOME_RATIO = AMT_ANNUITY / AMT_INCOME_TOTAL (Debt burden)
EXT_SOURCES_MEAN = (EXT_SOURCE_1 + EXT_SOURCE_2 + EXT_SOURCE_3) / 3 (Bureau composite)
Weight of Evidence ($\text{WoE}$) & Information Value ($\text{IV}$): Features evaluated with monotonic binning. Only variables with $\text{IV} \ge 0.02$ retained; multicollinear features pruned using Variance Inflation Factor ($\text{VIF} < 5.0$).

2. Model Benchmark & Statistical Performance

Model Architecture	Cross-Validation Strategy	AUC-ROC	Gini Coefficient	KS-Statistic	Brier Score Loss	Optimal Cut-off PD
Baseline Logistic Regression (WoE)	5-Fold Stratified	0.7420	0.4840	36.12%	0.0712	0.160
Random Forest Baseline	5-Fold Stratified	0.7315	0.4630	34.80%	0.0734	0.155
XGBoost Classifier	5-Fold Stratified	0.7762	0.5524	41.20%	0.0682	0.148
Champion LightGBM (Optuna Tuned)	5-Fold Stratified	0.7854	0.5708	43.17%	0.0664	0.145 (Policy: 0.180)

Note

Key Metric Definitions:

$\text{Gini} = 2 \times \text{AUC} - 1 = 2(0.7854) - 1 = \mathbf{0.5708}$
Kolmogorov-Smirnov ($\text{KS}$): $\text{KS} = \max |F_{\text{Bad}}(s) - F_{\text{Good}}(s)| = \mathbf{43.17%}$ ($p\text{-value} = 0.0000$), exceeding top-tier bank discrimination standards ($\ge 40%$).
Bad Capture Rate: Top $30%$ highest-risk applicants capture $72.5%$ of all portfolio defaults.

3. FICO Point Scorecard Mathematical Calibration

Model-predicted default probabilities ($\text{PD}$) are converted into a standard FICO Point Scorecard ($300\text{--}850$):

$$\text{Odds} = \frac{1 - \text{PD}}{\max(\text{PD}, 10^{-6})}$$

$$\text{Factor} = \frac{\text{PDO}}{\ln(2)} = \frac{20}{\ln(2)} \approx 28.853904$$

$$\text{Offset} = \text{Base Score} - \left(\text{Factor} \times \ln(\text{Base Odds})\right) = 600 - (28.853904 \times \ln(50)) \approx 487.12$$

$$\text{Credit Score} = \text{clip}\left(487.12 + 28.8539 \times \ln(\text{Odds}), 300, 850\right)$$

Scorecard Calibration Parameters

Base Score: $600$ Points at Base Odds of $50:1$
Points to Double Odds ($\text{PDO}$): $20$ Points
Policy Cut-off Threshold: $\text{PD} \le 0.180 \implies \text{Score} \ge 531 \text{ Points}$
Population Distribution: Mean = $566.01$, Median = $568.00$, $\text{IQR} = [549, 584]$

Model Explainability (XAI) & Regulatory Compliance (FCRA / ECOA)

Under the Fair Credit Reporting Act (FCRA) and Equal Credit Opportunity Act (ECOA), lenders issuing an adverse credit decision must provide the applicant with specific, verifiable reason codes explaining the rejection.

+-----------------------------------------------------------------------------------+
|                     OFFICIAL ADVERSE ACTION STATEMENT                            |
+-----------------------------------------------------------------------------------+
| Application ID: #281935                 Status: REJECTED                          |
| Model Score: 483 Points (Deep Subprime) Predicted Default Probability: 64.82%     |
+-----------------------------------------------------------------------------------+
| Key Risk Drivers (TreeSHAP Game-Theoretic Attribution):                           |
|  [RC_01] Low External Credit Bureau Rating                                        |
|         -> Description: External credit bureau score is below standard cut-off.   |
|         -> Remedial: Maintain on-time debt repayment history for >= 6 months.     |
|                                                                                   |
|  [RC_04] High Monthly Debt-to-Loan Ratio                                          |
|         -> Description: Monthly installment obligation is too high relative to credit.|
|         -> Remedial: Consider extending tenure to lower monthly installment.       |
|                                                                                   |
|  [RC_06] High Cash Flow Volatility                                                |
|         -> Description: Historical monthly cash flow volatility is elevated.      |
|         -> Remedial: Maintain stable average account balance reserves >= 3-6 mos. |
+-----------------------------------------------------------------------------------+

1. Adverse Action Reason Codes Mapping

Code	Indicator	Primary Trigger	Regulatory Description	Actionable Remediation
`RC_01`	`EXT_SOURCES_MEAN`	$< 0.35$	External credit bureau rating below standard cut-off	Maintain on-time debt repayment history for $\ge 6$ consecutive months
`RC_02`	`BURO_DAYS_CREDIT_MAX`	$> -30\text{ days}$	Excessive recent credit inquiries in a short timeframe	Pause new credit applications for 3–6 months to avoid debt stacking
`RC_03`	`EXT_SOURCE_3`	$< 0.30$	Low external score provider 3 rating	Establish positive trade lines and verify credit bureau records
`RC_04`	`CREDIT_TERM`	$> 0.055$	Monthly installment burden is too high relative to credit	Extend loan tenure to reduce the monthly annuity burden
`RC_05`	`AMT_ANNUITY`	High	Monthly payment exceeds disposable income ceiling	Reduce loan amount or increase down payment
`RC_06`	`cash_flow_volatility`	$> 0.30$	Elevated month-to-month cash flow volatility	Maintain stable average account balance reserves for 3–6 months
`RC_GEN_1`	`COMPOSITE_SCORE`	$\text{PD} > 0.180$	Overall risk score below minimum bank cut-off	Reduce loan exposure or add creditworthy co-signers

2. Actionable Counterfactual Recourse (DiCE Optimization)

Rather than simply rejecting an applicant, the engine executes constrained optimization across mutable features (AMT_CREDIT, AMT_ANNUITY, AMT_GOODS_PRICE) while freezing immutable demographic features (DAYS_BIRTH, CODE_GENDER):

$$\min_{\mathbf{x}^{\ast}} \text{dist}(\mathbf{x}, \mathbf{x}^{\ast}) \quad \text{subject to} \quad f(\mathbf{x}^{\ast}) \le \text{Cut-off PD } (0.180)$$

Recourse Action Matrix

Loan Principal Reduction: Reduce requested loan amount by $20%$ $\implies \text{Simulated PD drops from } 64.8% \text{ to } 55.1%$.
Tenure Extension: Extend tenure by $25%$ to reduce monthly annuity $\implies \text{Simulated PD drops to } 48.6%$.
Combined Down Payment & Tenor: $20%$ Down payment $+ 6$ months on-time payment track record $\implies \text{Simulated PD drops to } \mathbf{12.4%}$ ($\mathbf{APPROVED}$).

Model Governance, PSI & Macroeconomic Stress Testing (ICAAP)

1. Population Stability Index ($\text{PSI}$) Formulation

The monitoring pipeline continuously evaluates score distribution shifts between the baseline training cohort ($E_i$) and incoming production streams ($A_i$):

$$\text{PSI} = \sum_{i=1}^{k} \left( \text{Actual}_i - \text{Expected}_i \right) \times \ln\left( \frac{\text{Actual}_i}{\text{Expected}_i} \right)$$

     PSI < 0.10          0.10 <= PSI <= 0.25               PSI > 0.25
 ┌─────────────────┐    ┌────────────────────┐    ┌───────────────────────────┐
 │ 🟢 GREEN TIER   │    │  🟡 YELLOW TIER    │    │      🔴 RED TIER          │
 │ Population is   │    │  Moderate Drift    │    │ Significant Shift         │
 │ Stable; proceed │ -> │  Increase audit    │ -> │ Trigger Mandatory Model   │
 │ with normal ops │    │  frequency         │    │ Retraining & Policy Review│
 └─────────────────┘    └────────────────────┘    └───────────────────────────┘

2. Macro Stress Testing & Capital Adequacy (ICAAP Leaderboard)

Under the Internal Capital Adequacy Assessment Process (ICAAP), the bank estimates Expected Loss ($\text{EL} = \text{EAD} \times \text{PD} \times \text{LGD}$) and computes Required Capital Buffer Deltas under severe macroeconomic shocks:

Scenario Name	Description	Avg Stressed PD	Simulated NPL	Expected Loss (M THB)	Capital Buffer Delta	PSI Index	Governance Action
1. Baseline Scenario	Normal macroeconomic climate	8.07%	2.85%	348.0 M	0.0 M	0.024 (🟢 Stable)	Standard Basel III monitoring
2. Gig-Worker Shock	$-30%$ cash flow stability in gig-economy	11.45%	4.20%	418.9 M	+70.9 M	0.142 (🟡 Warning)	Tighten verification for variable-income applicants
3. Stagflation & Rate Hike	Inflation $+4.5%$, Policy rate $+200\text{ bps}$	14.82%	5.57%	458.3 M	+110.3 M	0.285 (🔴 Retrain)	Mandatory capital buffer injection & policy cut-off tightening

Production Application Stack & Key Features

Technology Stack Overview

+------------------------------------------------------------------------------------+
|  FRONTEND (Client Tier)                                                            |
|  * Next.js 16 (React 19, App Router, TypeScript 5)                                 |
|  * Tailwind CSS 4.0 (Custom FinTech Light Theme, Slate/Zinc Palette)               |
|  * Lucide Icons & Responsive Widescreen Layout (Max 1440px Container)              |
|  * Global Bilingual Context (🇹🇭 Thai | 🇬🇧 English Instant Switcher)             |
|  * Interactive Onboarding Modal & Searchable Financial Glossary                    |
+-----------------------------------------▲------------------------------------------+
                                          │ REST API JSON
+-----------------------------------------▼------------------------------------------+
|  BACKEND (Microservices Tier)                                                      |
|  * FastAPI 0.115 (High-throughput async Python 3.13 ASGI Engine)                   |
|  * Pydantic V2 Schema Validation & Error Normalization                             |
|  * LightGBM 5-Fold Stratified Ensemble Inference Engine                            |
|  * Joblib Artifact Caching & Strict Categorical Schema Locking                     |
|  * SQLAlchemy 2.0 ORM & SQLite WAL-Mode Audit Trail Database                       |
+------------------------------------------------------------------------------------+

Dashboard Tabs Architecture

1. 🎯 Tab 1: Portfolio Credit Policy & Net Profit Simulator

Interactive Policy Parameters: Real-time sliders for Cut-off PD Threshold (%), Average Loan Interest Rate (%), and Loss Given Default - LGD (%).
4 Big KPI Cards: Approval Rate ($% / \text{Count}$), Expected Default Rate ($\text{EDR } %$), Approved Exposure (M THB), and Simulated Net Profit (M THB).
Strategic Policy Recommendations: Dynamic business rule guidance (Conservative Capital Preservation, Balanced Growth, Aggressive Expansion).

2. 👤 Tab 2: Individual Underwriting & Recourse Engine

Applicant Risk Profiling: Input sliders with 1-click presets (👤 Prime Profile, ⚠️ Subprime, 💼 Gig-Worker).
Single-Row Decision Verdict: High-contrast verdict badge (🟢 APPROVED / 🔴 REJECTED) aligned with Predicted Default Risk ($\text{PD } %$).
Point Scorecard Gauge: Semi-circular SVG gauge displaying calibrated FICO Score ($300\text{--}850$) with risk tiers (Super Prime to Deep Subprime).
Adverse Action Notice & Recourse: Full-width reason code cards with actionable counterfactual recourse options.
Regulatory Audit Trail: Live SQLite audit table with timestamps, input payloads, scores, and underwriting decisions.

3. ⚡ Tab 3: Macroeconomic Stress Testing & PSI Tracker

Widescreen Scenario Selector: Instant simulation across Baseline, Gig-Worker Shock, and Stagflation scenarios.
ICAAP Benchmark Comparison: Comparison cards showing Stressed Average PD, Simulated NPL, and Required Capital Buffer Delta ($+\text{M THB}$).
Population Stability Index (PSI) Drift Monitor: Status badges and governance action directives.

Quick Start & Local Setup Guide

Prerequisites

Python: Version 3.10+ (Recommended: 3.13)
Node.js: Version 18.0+ (Recommended: v24.x)
Package Managers: pip and npm

Step 1: Clone the Repository

git clone https://github.com/svkhun/Credit-Risk-Scoring-Explainable-AI.git
cd "Credit Risk Scoring & Explainable AI"

Step 2: Backend Setup & Launch (Terminal 1)

# 1. Create and activate a Python virtual environment
python -m venv .venv

# On Windows:
.venv\Scripts\activate
# On macOS / Linux:
source .venv/bin/activate

# 2. Install required Python dependencies
pip install -r backend/requirements.txt

# 3. Start the FastAPI microservice
python -m uvicorn backend.app.main:app --host 0.0.0.0 --port 8000 --reload

Backend Service URL: http://localhost:8000
Interactive Swagger Documentation: http://localhost:8000/docs
Health Check Endpoint: http://localhost:8000/health

Step 3: Frontend Setup & Launch (Terminal 2)

# 1. Navigate to the frontend directory
cd frontend

# 2. Install Node dependencies
npm install

# 3. Launch the Next.js development server
npm run dev

Web Dashboard URL: http://localhost:3000

Step 4: Verify API Endpoints with cURL

# Test Individual Underwriting Assessment
curl -X POST http://localhost:8000/api/v1/score \
  -H "Content-Type: application/json" \
  -d '{
    "AMT_CREDIT": 650000.0,
    "AMT_ANNUITY": 42000.0,
    "AMT_GOODS_PRICE": 600000.0,
    "DAYS_BIRTH": -9125,
    "CODE_GENDER": "M",
    "EXT_SOURCES_MEAN": 0.16,
    "EXT_SOURCE_3": 0.18,
    "BURO_DAYS_CREDIT_MAX": -12.0,
    "cash_flow_volatility": 0.48
  }'

# Test Policy Profit Simulation
curl -X POST http://localhost:8000/api/v1/simulate/policy \
  -H "Content-Type: application/json" \
  -d '{
    "cutoff_pd": 18.0,
    "interest_rate": 15.0,
    "lgd": 45.0,
    "total_portfolio_size": 15000,
    "avg_loan_amount": 600000.0
  }'

Repository Directory Structure

Credit-Risk-Scoring-Explainable-AI/
├── backend/                                # FastAPI Microservices Application
│   ├── app/
│   │   ├── __init__.py
│   │   ├── main.py                         # FastAPI routes & CORS orchestration
│   │   ├── database.py                     # SQLAlchemy session & SQLite engine
│   │   ├── models_db.py                    # SQLite Audit Trail ORM models
│   │   ├── schemas.py                      # Pydantic v2 validation schemas
│   │   └── services.py                     # LightGBM scoring, scorecard & XAI engine
│   └── requirements.txt                    # Backend dependencies
│
├── frontend/                               # Next.js 16 (React 19) Application
│   ├── src/
│   │   ├── app/
│   │   │   ├── globals.css                 # Clean Light-theme Tailwind layer styles
│   │   │   ├── layout.tsx                  # Root layout with fonts & LanguageProvider
│   │   │   └── page.tsx                    # Widescreen 1440px multi-tab orchestrator
│   │   ├── components/
│   │   │   ├── Navbar.tsx                  # Light theme header with TH/EN switch
│   │   │   ├── TabPolicySimulator.tsx      # Tab 1: Net Profit & Policy Simulator
│   │   │   ├── TabUnderwriting.tsx         # Tab 2: Individual Underwriting
│   │   │   ├── TabStressTesting.tsx        # Tab 3: Macro Stress Testing & PSI
│   │   │   ├── ScoreGauge.tsx              # SVG FICO Point Scorecard gauge (300-850)
│   │   │   ├── AdverseNotice.tsx           # FCRA / ECOA Adverse Action Notice cards
│   │   │   ├── RecourseTable.tsx           # Actionable counterfactual recourse table
│   │   │   ├── AuditTrail.tsx              # Regulatory SQLite audit trail table
│   │   │   ├── WelcomeModal.tsx            # First-visit onboarding modal
│   │   │   └── GlossaryModal.tsx           # Searchable financial & variable glossary
│   │   ├── context/
│   │   │   └── LanguageContext.tsx         # Bilingual Context with localStorage persistence
│   │   └── lib/
│   │       ├── api.ts                      # Typed client for all 4 FastAPI endpoints
│   │       └── translations.ts             # Complete bilingual dictionary & reason code mappings
│   ├── package.json
│   ├── tsconfig.json
│   └── next.config.ts
│
├── notebooks/                              # Quantitative Research & Modeling Pipeline
│   ├── 01_eda_imputation.ipynb             # Exploratory Data Analysis & Imputation
│   ├── 02_feature_engineering.ipynb        # WoE Binning, IV analysis & Ratios
│   ├── 03_baseline_modeling.ipynb          # 5-Fold LightGBM, Optuna tuning & Calibration
│   ├── 04_explainability_recourse.ipynb    # TreeSHAP & DiCE Counterfactuals
│   └── 05_model_governance_monitoring.ipynb# PSI/CSI Data Drift & ICAAP Stress Testing
│
├── models/                                 # Pre-trained Model Artifacts
│   ├── champion_lightgbm_5folds.pkl        # Serialized 5-fold LightGBM ensemble
│   └── oof_predictions.npy                 # Out-of-fold validation probability array
│
├── data/                                   # Data Assets
│   ├── raw/                                # Raw Home Credit & Bureau CSV files
│   └── processed/
│       └── train_engineered.parquet        # Reference engineered dataset schema
│
├── credit_risk_platform.db                 # SQLite WAL regulatory audit log database
└── README.md                               # Project documentation

Regulatory Compliance & Model Governance

This platform adheres to global credit risk modeling and algorithmic governance standards:

Basel II / Basel III Framework: Internal Ratings-Based (IRB) approach for Probability of Default ($\text{PD}$) estimation and capital requirement calculations.
Fair Credit Reporting Act (FCRA) & Equal Credit Opportunity Act (ECOA): Algorithmic transparency providing Top-4 Adverse Action Reason Codes for adverse underwriting decisions.
Internal Capital Adequacy Assessment Process (ICAAP): Forward-looking macroeconomic stress testing assessing solvency under extreme economic shocks.
EBA Guidelines on Machine Learning in Credit Risk: Preserving feature monotonicity, stability audits ($\text{PSI} < 0.10$), and eliminating demographic proxy bias.

License & Author

Project Lead / Author: Quantitative Risk & AI Engineering Team
Repository: svkhun/Credit-Risk-Scoring-Explainable-AI
License: MIT Open Source License

Navigation Menu

Repository files navigation

Credit Risk Scoring & Explainable AI Intelligence Platform

Executive Summary

End-to-End System Architecture

Quantitative Risk Modeling & Machine Learning Core

1. Data Engineering & Feature Pipeline

2. Model Benchmark & Statistical Performance

3. FICO Point Scorecard Mathematical Calibration

Scorecard Calibration Parameters

Model Explainability (XAI) & Regulatory Compliance (FCRA / ECOA)

1. Adverse Action Reason Codes Mapping

2. Actionable Counterfactual Recourse (DiCE Optimization)

Recourse Action Matrix

Model Governance, PSI & Macroeconomic Stress Testing (ICAAP)

1. Population Stability Index ($\text{PSI}$) Formulation

2. Macro Stress Testing & Capital Adequacy (ICAAP Leaderboard)

Production Application Stack & Key Features

Technology Stack Overview

Dashboard Tabs Architecture

1. 🎯 Tab 1: Portfolio Credit Policy & Net Profit Simulator

2. 👤 Tab 2: Individual Underwriting & Recourse Engine

3. ⚡ Tab 3: Macroeconomic Stress Testing & PSI Tracker

Quick Start & Local Setup Guide

Prerequisites

Step 1: Clone the Repository

Step 2: Backend Setup & Launch (Terminal 1)

Step 3: Frontend Setup & Launch (Terminal 2)

Step 4: Verify API Endpoints with cURL

Repository Directory Structure

Regulatory Compliance & Model Governance

License & Author

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
.vscode		.vscode
app		app
backend/app		backend/app
frontend		frontend
notebooks		notebooks
.gitignore		.gitignore
README.md		README.md
credit_risk_platform.db		credit_risk_platform.db
requirements.txt		requirements.txt

Latest commit

History

Folders and files

Repository files navigation

Credit Risk Scoring & Explainable AI Intelligence Platform

Executive Summary

End-to-End System Architecture

Quantitative Risk Modeling & Machine Learning Core

1. Data Engineering & Feature Pipeline

2. Model Benchmark & Statistical Performance

3. FICO Point Scorecard Mathematical Calibration

Scorecard Calibration Parameters

Model Explainability (XAI) & Regulatory Compliance (FCRA / ECOA)

1. Adverse Action Reason Codes Mapping

2. Actionable Counterfactual Recourse (DiCE Optimization)

Recourse Action Matrix

Model Governance, PSI & Macroeconomic Stress Testing (ICAAP)

1. Population Stability Index ($\text{PSI}$) Formulation

2. Macro Stress Testing & Capital Adequacy (ICAAP Leaderboard)

Production Application Stack & Key Features

Technology Stack Overview

Dashboard Tabs Architecture

1. 🎯 Tab 1: Portfolio Credit Policy & Net Profit Simulator

2. 👤 Tab 2: Individual Underwriting & Recourse Engine

3. ⚡ Tab 3: Macroeconomic Stress Testing & PSI Tracker

Quick Start & Local Setup Guide

Prerequisites

Step 1: Clone the Repository

Step 2: Backend Setup & Launch (Terminal 1)

Step 3: Frontend Setup & Launch (Terminal 2)

Step 4: Verify API Endpoints with cURL

Repository Directory Structure

Regulatory Compliance & Model Governance

License & Author

About

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages