Professional dataset card: 5K synthetic fraud benchmark

README.md

@@ -1,76 +1,181 @@
 ---
 license: apache-2.0
-dataset_info:
-  features:
-  - name: transaction_id
-    dtype: string
-  - name: customer_id
-    dtype: string
-  - name: transaction_hour
-    dtype: int64
-  - name: account_age_days
-    dtype: int64
-  - name: previous_chargebacks
-    dtype: int64
-  - name: merchant_category
-    dtype: string
-  - name: transaction_country
-    dtype: string
-  - name: device_type
-    dtype: string
-  - name: is_international
-    dtype: int64
-  - name: is_high_risk_merchant
-    dtype: int64
-  - name: transaction_amount
-    dtype: float64
-  - name: transaction_velocity_1h
-    dtype: int64
-  - name: transaction_velocity_24h
-    dtype: int64
-  - name: avg_transaction_amount_30d
-    dtype: float64
-  - name: risk_label
-    dtype: int64
-  splits:
-  - name: train
-    num_bytes: 679378
-    num_examples: 5000
-  download_size: 171991
-  dataset_size: 679378
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
----
+language:
+- en
 task_categories:
 - tabular-classification
 tags:
 - fraud-detection
+- tabular
+- financial
 - trustworthy-ai
-- lead-ai
-- financial-ai
+- responsible-ai
+- explainable-ai
+- xai
+- synthetic-data
+- lead-ai-labs
+size_categories:
+- 1K<n<10K
 ---
 
 # Lead.AI Fraud Detection Dataset
 
-This dataset simulates real-world credit card transactions for fraud detection research and AI model development.
+**5,000-row synthetic tabular dataset for fraud detection research, XAI benchmarking, and predictive model training**
+
+> Published by [Lead.AI Labs](https://www.lead-ai.us)  
+> Author: [Arun Kumar Gharami](https://huggingface.co/arun-gharami) | [Google Scholar](https://scholar.google.com/citations?user=uy4i5soAAAAJ&hl=en)
+
+---
+
+## Dataset Summary
+
+This dataset contains 5,000 synthetic financial transactions labeled as fraud (1) or normal (0).
+It was designed to support research and prototyping in explainable fraud detection, featuring
+a balanced mix of behavioral, temporal, and merchant-level risk signals.
+
+> ⚠️ **Synthetic Data Disclaimer:** This dataset is **entirely synthetic**. It was generated
+> algorithmically to simulate plausible transaction fraud patterns for research and educational
+> use. It contains **no real customer data, no real financial transactions, and no personally
+> identifiable information (PII)**. It must not be treated as representative of any real-world
+> fraud distribution.
+
+---
+
+## Supported Tasks
+
+- Binary fraud classification (`risk_label`: 0 = normal, 1 = fraud)
+- Explainable AI (XAI) feature attribution research
+- Imbalanced classification benchmarking
+- Financial risk signal analysis
+
+---
+
+## Data Fields
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `transaction_id` | string | Unique transaction identifier |
+| `customer_id` | string | Unique customer identifier |
+| `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 |
+| `merchant_category` | string | Merchant type (online_services, grocery, restaurant, electronics, travel, fuel, fashion) |
+| `transaction_country` | string | Country code (US, UK, CA, AU, IN) |
+| `device_type` | string | Device used (mobile, desktop, tablet) |
+| `is_international` | int | 1 = international transaction |
+| `is_high_risk_merchant` | int | 1 = high-risk merchant category |
+| `transaction_velocity_1h` | int | Transaction count in last 1 hour |
+| `transaction_velocity_24h` | int | Transaction count in last 24 hours |
+| `avg_transaction_amount_30d` | float | 30-day average transaction amount |
+| `risk_label` | int | **Target variable** — 0 = normal, 1 = fraud |
+
+---
 
-## Features
+## Example Rows
+
+| transaction_id | transaction_amount | transaction_hour | account_age_days | is_international | risk_label |
+|----------------|-------------------|------------------|------------------|-----------------|------------|
+| TXN0000047 | 922.01 | 1 | 12 | 1 | 1 (fraud) |
+| TXN0000094 | 149.22 | 14 | 340 | 0 | 1 (fraud) |
+| TXN0000099 | 70.30 | 9 | 892 | 0 | 1 (fraud) |
+| TXN0001234 | 45.80 | 11 | 720 | 0 | 0 (normal) |
+
+---
 
-- Time: Transaction timestamp
-- V1–V28: Anonymized features (PCA-based)
-- Amount: Transaction value
-- Class: Fraud label (0 = normal, 1 = fraud)
+## How to Load
 
-## Use Cases
+### With `datasets` (Hugging Face)
+```python
+from datasets import load_dataset
 
-- Fraud detection modeling
-- Explainable AI (XAI)
-- Financial risk analysis
+ds = load_dataset("arun-gharami/lead-ai-fraud-detection-dataset")
+df = ds["train"].to_pandas()
+print(df.head())
+```
+
+### With `pandas`
+```python
+import pandas as pd
+
+url = "hf://datasets/arun-gharami/lead-ai-fraud-detection-dataset/data/train-00000-of-00001.parquet"
+df = pd.read_parquet(url)
+print(df.shape)
+print(df["risk_label"].value_counts())
+```
+
+### With DuckDB (Parquet)
+```python
+import duckdb
+
+result = duckdb.query("""
+    SELECT risk_label, COUNT(*) as count, AVG(transaction_amount) as avg_amount
+    FROM read_parquet('hf://datasets/arun-gharami/lead-ai-fraud-detection-dataset/data/train-00000-of-00001.parquet')
+    GROUP BY risk_label
+""").df()
+print(result)
+```
+
+---
 
-## Author
+## Dataset Splits
+
+| Split | Rows |
+|-------|------|
+| train | 5,000 |
+
+---
+
+## Intended Use
+
+- Training and evaluating fraud detection classifiers
+- XAI research: SHAP, LIME, feature attribution on tabular financial data
+- Academic benchmarking and reproducibility studies
+- Prototyping fraud detection dashboards and APIs
+
+## Non-Intended Use
+
+- Training production fraud systems deployed on real customer data
+- Academic claims about real-world fraud rates or distributions
+- Any use involving actual customer financial data
+
+---
+
+## Bias and Fairness Note
+
+This dataset was synthetically generated and may encode assumptions about the relationship
+between demographic signals (country, device type) and fraud likelihood that do not reflect
+real-world distributions. Users should audit for proxy discrimination before using any
+model trained on this data in a context involving real individuals.
+
+---
+
+## Privacy Note
+
+This dataset contains **no real PII**. All transaction IDs, customer IDs, amounts, and
+behavioral features are synthetically generated.
+
+---
+
+## Citation
+
+```bibtex
+@misc{gharami2024frauddataset,
+  author       = {Arun Kumar Gharami},
+  title        = {Lead.AI Fraud Detection Dataset: Synthetic Tabular Benchmark for XAI Research},
+  year         = {2024},
+  publisher    = {Hugging Face},
+  howpublished = {\url{https://huggingface.co/datasets/arun-gharami/lead-ai-fraud-detection-dataset}}
+}
+```
+
+---
+
+## License
+
+Apache 2.0
+
+---
 
-Arun Kumar Gharami  
-AI Researcher | Trustworthy ML
+*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)