---
license: mit
language:
- en
tags:
- fraud-detection
- binary-classification
- tabular-classification
- explainable-ai
- xai
- trustworthy-ai
- responsible-ai
- finance
- scikit-learn
- lead-ai-labs
- synthetic-data
pipeline_tag: tabular-classification
library_name: sklearn
---

# Lead.AI Fraud Detection Model

**Binary transaction fraud classifier with transparent decision reasoning and XAI explainability**

> Part of the [Lead.AI Labs](https://www.lead-ai.us) Trustworthy AI Suite  
> Author: [Arun Kumar Gharami](https://huggingface.co/arun-gharami) | [Google Scholar](https://scholar.google.com/citations?user=uy4i5soAAAAJ&hl=en)

---

## Model Overview

Lead.AI Fraud Detection Model is a binary tabular classifier that labels individual financial
transactions as **Fraud** or **Safe** using behavioral and financial features. Alongside each
prediction, the model surfaces a risk score and a plain-language explanation of the key
factors driving the decision — making AI-assisted fraud triage transparent and auditable.

This is the foundational classifier in the Lead.AI Fraud Detection system. For the extended
three-tier risk scorer (Low / Medium / High), see
[lead-ai-fraud-shield](https://huggingface.co/arun-gharami/lead-ai-fraud-shield).

The model is designed for **educational, research, prototyping, and demonstration** contexts.

---

## Intended Use

**Appropriate uses:**
- Research and experimentation with explainable fraud detection classifiers
- Prototyping fraud risk dashboards, triage queues, and investigation tools
- Educational demonstrations of XAI in financial ML contexts
- Interactive demos (see [Fraud Detection XAI Demo](https://huggingface.co/spaces/arun-gharami/fraud-detection-xai-demo))

**Not intended for:**
- Production fraud enforcement or automated transaction blocking
- Sole basis for financial, legal, or fraud-related decisions about individuals
- Deployment in regulated financial systems without proper validation and compliance review

---

## Features

| Feature | Type | Description |
|---------|------|-------------|
| `transaction_amount` | float | Transaction value in currency units |
| `transaction_hour` | int | Hour of transaction (0–23) |
| `account_age_days` | int | Days since account creation |
| `previous_chargebacks` | int | Historical chargeback count |
| `transaction_velocity_1h` | int | Number of transactions in last 1 hour |
| `transaction_velocity_24h` | int | Number of transactions in last 24 hours |
| `is_international` | int | Binary flag: 1 = international transaction |
| `is_high_risk_merchant` | int | Binary flag: 1 = high-risk merchant category |
| `customer_risk_score` | float | Aggregated customer risk indicator (0.0–1.0) |

---

## Output Format

```json
{
  "prediction": "Fraud",
  "risk_score": 0.91,
  "explanation": "High transaction velocity in the past hour, new account, and international flag are the primary risk drivers."
}
```

| Output | Values |
|--------|--------|
| `prediction` | `Fraud` / `Safe` |
| `risk_score` | Float 0.0–1.0 (fraud probability) |
| `explanation` | Natural-language explanation of top risk drivers |

---

## Example Usage

```python
import joblib
import pandas as pd

model = joblib.load("model/model.joblib")

sample = pd.DataFrame([{
    "transaction_amount": 850.0,
    "transaction_hour": 2,
    "account_age_days": 12,
    "previous_chargebacks": 2,
    "transaction_velocity_1h": 5,
    "transaction_velocity_24h": 18,
    "is_international": 1,
    "is_high_risk_merchant": 1,
    "customer_risk_score": 0.87
}])

prediction = model.predict(sample)
probability = model.predict_proba(sample)
print(prediction, probability)
```

---

## Interactive Demo

Try the model live in your browser:

🖥️ **[Fraud Detection XAI Demo](https://huggingface.co/spaces/arun-gharami/fraud-detection-xai-demo)**  
Enter a transaction, get a fraud/safe prediction, and see the SHAP-style explanation instantly.

---

## Explainability / XAI

This model surfaces a natural-language explanation with each prediction, identifying the top
features that drove the fraud or safe classification. The explanation is designed to support:

- Human review and investigation workflows
- Audit trail generation for flagged transactions
- User-facing transparency in decision-support tools
- Research into XAI methods for tabular financial data

---

## Training Data Note

This model was trained on the
[lead-ai-fraud-detection-dataset-v2](https://huggingface.co/datasets/arun-gharami/lead-ai-fraud-detection-dataset-v2)
— a **fully synthetic 100K-transaction dataset** generated to simulate plausible fraud
patterns. The dataset:
- Contains **no real customer financial data**
- Was generated for research, educational, and prototyping purposes
- May not capture the full diversity of real-world fraud distributions

---

## Responsible AI & Limitations

- **Synthetic training data:** Real-world performance is unknown and unvalidated
- **No regulatory certification:** Not reviewed or certified by any financial regulatory body
- **Binary framing:** Real fraud landscapes are probabilistic and contextual; a binary label
  is a simplification
- **False positive risk:** Innocent transactions may be flagged; human review is essential
- **Concept drift:** Fraud patterns evolve; models require periodic retraining

---

## Ethical Use Statement

> This model is intended for educational, research, prototyping, and decision-support use only.
> It **must not** be used as the sole basis for financial, legal, lending, or fraud enforcement
> decisions affecting individuals. Users are responsible for all validation, bias testing,
> legal review, and compliance requirements before any production deployment.

---

## Citation

```bibtex
@misc{gharami2024frauddetection,
  author       = {Arun Kumar Gharami},
  title        = {Lead.AI Fraud Detection Model: Binary Transaction Fraud Classifier with XAI},
  year         = {2024},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/arun-gharami/lead-ai-fraud-detection-model}}
}
```

---

## License

MIT — See [LICENSE](LICENSE) for details.

---

*Lead.AI Labs — Trustworthy AI Systems for Practical Business Intelligence*  
[lead-ai.us](https://www.lead-ai.us) · [GitHub](https://github.com/Arungharami) · [LinkedIn](https://www.linkedin.com/in/arunkgharami)