Durgesh111/Cifer-Fraud-Detection-Dataset-AF

---
license: apache-2.0
task_categories:
- tabular-classification
- feature-extraction
language:
- en
tags:
- fraud-detection
- finance
- federated-learning
- cifer
pretty_name: Cifer-Fraud-Detection-Dataset-AF
size_categories:
- 10M<n<100M
---

# 📊 Cifer Fraud Detection Dataset

## 🧠 Overview
The **Cifer-Fraud-Detection-Dataset-AF** is a high-fidelity, fully synthetic dataset created to support the development and benchmarking of privacy-preserving, federated, and decentralized machine learning systems in financial fraud detection.

This dataset draws structural inspiration from the **PaySim simulator,** which was built using aggregated mobile money transaction data from a real financial provider operating in 14+ countries. Cifer extends this format by scaling it to **21 million samples,** optimizing for **federated learning environments,** and validating performance against real-world datasets.

> ### Accuracy Benchmark:
> Cifer-trained models on this dataset reach **99.93% accuracy,** benchmarked against real-world fraud datasets with **99.98% baseline accuracy**—providing high-fidelity behavior for secure, distributed ML research.

---

## ⚙️ Generation Method
This dataset is **entirely synthetic** and was generated using **Cifer’s internal simulation engine,** trained to mimic patterns of financial behavior, agent dynamics, and fraud strategies typically observed in mobile money ecosystems.
- Based on the structure and simulation dynamics of PaySim
- Enhanced for multi-agent testing, federated partitioning, and async model training
- Includes realistic fraud flagging mechanisms and unbalanced label distributions

---

# 🧩 Data Structure
| Column Name | Description |
|------------------|-----------------------------------------------------------------------------|
| `step` | Unit of time (1 step = 1 hour); simulation spans 30 days (744 steps total) |
| `type` | Transaction type: CASH-IN, CASH-OUT, DEBIT, PAYMENT, TRANSFER |
| `amount` | Transaction value in simulated currency |
| `nameOrig` | Anonymized ID of sender |
| `oldbalanceOrg` | Sender’s balance before transaction |
| `newbalanceOrig` | Sender’s balance after transaction |
| `nameDest` | Anonymized ID of recipient |
| `oldbalanceDest` | Recipient’s balance before transaction (if applicable) |
| `newbalanceDest` | Recipient’s balance after transaction (if applicable) |
| `isFraud` | Binary flag: 1 if transaction is fraudulent |
| `isFlaggedFraud` | 1 if transaction exceeds a flagged threshold (e.g. >200,000) |

---

# 📁 File Organization
Total Rows: **21,000,000**
Split into 14 files for large-scale and federated learning scenarios:
- `Cifer-Fraud-Detection-Dataset-AF-part-1-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-2-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-3-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-4-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-5-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-6-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-7-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-8-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-9-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-10-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-11-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-12-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-13-14.csv` → 1.5M rows
- `Cifer-Fraud-Detection-Dataset-AF-part-14-14.csv` → 1.5M rows

Format: `.csv` (optionally `.parquet` or `.json` upon request)

---

# ✅ Key Features
- Fully synthetic and safe for public release
- Compatible with federated learning (cross-silo, async, or multi-agent)
- Ideal for privacy-preserving machine learning and robustness testing
- Benchmarkable against real-world fraud datasets
- Supports fairness evaluation via distribution-aware modeling

---

# 🔬 Use Cases
- Fraud detection benchmarking in decentralized AI systems
- Federated learning simulation (training, evaluation, aggregation)
- Model bias mitigation and fairness testing
- Multi-agent coordination and adversarial fraud modeling

---

# 📜 License
**Apache 2.0** — freely usable with attribution

---

# 🧾 Attribution & Citation
This dataset was generated and extended by Cifer AI, building on structural principles introduced by:

**E. A. Lopez-Rojas, A. Elmir, and S. Axelsson** <br>
*PaySim: A financial mobile money simulator for fraud detection.* <br>
28th European Modeling and Simulation Symposium – EMSS 2016

---