💱 FX Exposure Analytics & Anomaly Detection

A Python + Streamlit application for analyzing foreign exchange (FX) exposure, detecting anomalies using machine learning, and generating risk-based insights from ERP-style transaction data.

---

📌 Overview

This project simulates an enterprise FX exposure management workflow — the kind used in corporate treasury and finance operations to monitor currency risk across business units and geographies.

It includes:

• Synthetic FX exposure data generation (multi-currency, multi-entity)
• ML-based anomaly detection using Isolation Forest
• An interactive Streamlit dashboard for drill-down analysis
• Scored output datasets for downstream reporting

---

🗂️ Project Structure

FX_exposure/
├── app.py                              # Streamlit dashboard
├── train_fx_anomaly_model.py            # Train Isolation Forest anomaly model
├── test_fx_anomaly_model.py             # Unit tests for the anomaly model
├── fx_anomaly_pipeline.pkl             # Serialized trained ML pipeline
├── synthetic_fx_exposure_data.csv      # Synthetic FX exposure dataset
├── fx_exposure_with_anomaly_flags.csv  # Dataset with anomaly labels applied
├── june2025_erp_extract.csv            # Sample ERP extract (synthetic)
├── june2025_scored.csv                 # Scored output with anomaly flags
├── requirements.txt
└── .gitignore

---

⚙️ Key Features

Feature	Description
📊 Multi-Currency Exposure	Track exposure across USD, EUR, GBP, JPY, and more
🤖 Anomaly Detection	Isolation Forest ML model flags unusual FX transactions
🏢 Entity-Level Drill-Down	Breakdown by legal entity, business unit, or region
📉 Risk Scoring	Every transaction gets a scored anomaly probability
📊 Streamlit Dashboard	Interactive charts, filters, and exposure summaries
🧪 ERP-Style Data	Mimics real SAP/Oracle ERP extracts for realism

---

🚀 Getting Started

1. Clone the repository

git clone https://github.com/nmadagi/FX_exposure.git
cd FX_exposure

2. Install dependencies

pip install -r requirements.txt

3. Train the anomaly model (optional — pre-trained pkl included)

python train_fx_anomaly_model.py

4. Run the Streamlit dashboard

streamlit run app.py

---

🤖 ML Model: Isolation Forest

The anomaly detection pipeline uses Isolation Forest, an unsupervised algorithm well-suited for detecting outliers in high-dimensional financial data.

• Features include: notional amount, currency pair, entity, settlement date, transaction type
• Outputs an anomaly score and a binary flag (normal / anomalous)
• Trained pipeline serialized to fx_anomaly_pipeline.pkl for reuse

---

📦 Tech Stack

• Python 3.10+
• Streamlit — Interactive dashboard
• Scikit-learn — Isolation Forest anomaly detection
• Pandas / NumPy — Data wrangling
• Plotly — Visualizations
• Pickle — Model serialization

---

⚠️ Disclaimer

All data in this repository is fully synthetic. No real company, bank, or client data is used.

---

👤 Author

Nitin Madagi | GitHub | Portfolio

📄 License

This project is licensed under the MIT License.