Financial Risk & Customer Behaviour Intelligence

A customer analytics platform built on retail banking transaction data. The project covers data modelling, feature engineering, unsupervised segmentation, explainable risk scoring, supervised ML classification, and a Power BI dashboard.

---

Project Structure

financial-intelligence-platform/
├── notebooks/
│   ├── 1_data_preparation.ipynb           # data loading, synthetic dimensions, export
│   ├── 2_feature_engineering_segmentation.ipynb  # features, K-Means, risk scorecard
│   └── 3_risk_classification_models.ipynb  # Logistic Regression + Random Forest
├── data_raw/
│   └── creditcard.csv                     # Kaggle fraud dataset (not included in repo)
├── data_processed/
│   ├── FactTransactions.csv
│   ├── DimCustomer.csv
│   ├── DimAccount.csv
│   ├── DimMerchant.csv
│   ├── DimDate.csv
│   ├── CustomerFeatures.csv
│   ├── CustomerMonthlyFeatures.csv
│   ├── CustomerSegments.csv
│   └── RiskAlerts.csv
├── sql/
│   └── schema.sql                         # table definitions for all CSVs
├── reports/
│   └── executive_summary.md              # findings and recommendations
├── project.pbix                           # Power BI dashboard
└── requirements.txt

---

What This Project Does

1. Data Preparation (Notebook 1)

Loads the Kaggle credit card dataset and builds a realistic UK retail banking data model around it — customers, accounts, merchants, and channels — using synthetic but distribution-aware data generation.

2. Feature Engineering & Segmentation (Notebook 2)

Computes monthly and lifetime behavioural features per customer (spend volatility, ATM usage rate, online rate, cross-border activity, night transactions, discretionary spend).

Uses K-Means clustering to group customers into 5 behavioural segments:

• Digital Spenders
• Stable Essentials
• Cash-Heavy
• High-Volatility
• Cross-Border Lifestyle

Builds an explainable risk scorecard with reason codes (fraud history, spend spikes, etc.) and assigns each customer a risk tier: Low / Medium / High.

3. ML Classification (Notebook 3)

Trains Logistic Regression and Random Forest classifiers to predict at-risk customers. Random Forest achieves ~0.93 ROC-AUC. Feature importances show spend volatility and avg monthly spend are the strongest predictors.

4. Power BI Dashboard

Interactive dashboard with:

• Portfolio-level KPIs (total customers, risk distribution, fraud alerts)
• Segment breakdown and behaviour mapping
• Operational risk queue (Medium + High tier customers)
• Customer 360 drill-through (individual profiles with risk drivers)

---

How to Run

1. Clone the repo
2. Download creditcard.csv from Kaggle and place it in data_raw/
3. Install dependencies: pip install -r requirements.txt
4. Run the notebooks in order (1 → 2 → 3)
5. Open project.pbix in Power BI Desktop

---

Tech Stack

Area	Tools
Data processing	Python, Pandas, NumPy
Machine learning	Scikit-learn
Visualisation (code)	Matplotlib, Seaborn
Dashboard	Power BI
Schema	SQL

---

Dataset

Kaggle Credit Card Fraud Detection dataset — 284,807 transactions with PCA-transformed features and a fraud label. https://www.kaggle.com/datasets/mlg-ulb/creditcardfraud