Build sharper fraud models with better data.
Synthetic transaction datasets for fraud model development. Explore labeled payment-fraud scenarios, inspect the verified schema, and evaluate a representative sample before purchasing.
CSV + Parquet
Labeled edge cases
Commercial rights
| tx_id | amount | is_fraud | type | risk |
|---|---|---|---|---|
| tx_9012841 | €489.50 | 1 | velocity_attack | 0.92 |
| tx_9012842 | €12.00 | 0 | none | 0.05 |
| tx_9012843 | €1250.00 | 1 | impossible_travel | 0.88 |
| tx_9012844 | €45.20 | 0 | none | 0.12 |
01 / Dataset card
FraudData-Payment-v1
A public technical specification for reproducible model development.
Online payment fraud (card-not-present)
500,000 transactions
32 columns
1.85% · 9,250 labels
90 consecutive days
CSV + Apache Parquet
Core feature schema
| Field | Type | Role | Description |
|---|---|---|---|
| transaction_id | string | ID | Unique transaction hash |
| amount | float64 | Feature | Transaction value in EUR |
| is_fraud | int64 | Label | 0 = legitimate, 1 = fraud |
| fraud_type | string | Metadata | Scenario tag |
| risk_score | float64 | Feature | Device rating from 0.0 to 1.0 |
02 / Quality controls
Reliable by design. Honest about limits.
Data integrity is verified using automated relational and temporal checks. We declare explicit limitations rather than claiming artificial perfection.
No missing mandatory values or ID collisions.
User, card, and device references stay in valid ranges.
Chronological sequences avoid future information leakage.
Post-event resolution indicators stay out of training columns.
03 / Licensing
Start with the data you need.
One-time purchasing with transparent single-organization commercial rights. The free download is a 12-row CSV preview archive for technical evaluation.
04 / Quickstart
Go from download to split in minutes.
Load the dataset and execute a temporal train/test split in Python.
import pandas as pd
# The free download is a lightweight 12-row technical preview.
df = pd.read_csv('sample_fraud_dataset.csv')
df['timestamp'] = pd.to_datetime(df['timestamp'], utc=True)
# Separate predictors and label; exclude scenario and temporal leakage fields.
X = df.drop(columns=['is_fraud', 'timestamp', 'fraud_type'])
y = df['is_fraud']
# Temporal Train / Test Split (80/20)
cutoff = df['timestamp'].quantile(0.80)
train_mask = df['timestamp'] <= cutoff
X_train, y_train = X[train_mask], y[train_mask]
X_test, y_test = X[~train_mask], y[~train_mask]
print(f"Train set: {len(X_train):,} rows | Test set: {len(X_test):,} rows")Legal / Terms
Terms & Conditions
FraudData AI provides synthetic datasets for lawful model development, testing, and research. By downloading or purchasing a dataset, you agree to use it only for lawful purposes and to validate it before relying on any result. Paid data may be used by one purchasing legal organization, including its employees and contractors acting on its behalf, but may not be resold, sublicensed, publicly redistributed, or presented as live, personally sourced, or production-derived data. The service and datasets are provided as-is; to the fullest extent permitted by law, FraudData AI disclaims warranties of fitness, uninterrupted availability, and production performance. You are responsible for access controls, compliance with applicable law, and all decisions made using the datasets.
Legal / License
Single-Organization B2B Commercial License
Paid plans grant a perpetual, non-exclusive, non-transferable license to one named legal organization for internal analytics, research, training, evaluation, and product development. The license covers that organization and its personnel or service providers acting under written confidentiality obligations; it does not cover affiliates, clients, portfolio companies, or a parent or sister company unless separately licensed. Raw files, extracts, labels, credentials, and substantially similar derived datasets may not be sold, sublicensed, published, shared outside the licensed organization, or used to create a competing dataset product. Ownership of the underlying dataset and all rights not expressly granted remain with FraudData AI.
Legal / Privacy
Privacy Policy
Our datasets are 100% synthetic and are generated without real personal data, personally identifiable information, or records intended to represent real individuals. We apply data-minimisation, purpose-limitation, access-control, and retention principles consistent with GDPR. We collect only the business and contact information needed to process purchases, provide support, secure the service, and meet legal obligations; we do not sell personal data. Please contact support@ai-datasets.uk for privacy requests or questions. You remain responsible for assessing your own downstream processing and compliance obligations.
Legal / Refunds
Refund Policy
Because purchased datasets are digital goods delivered immediately, you expressly request immediate performance and acknowledge that the applicable cancellation right may end once delivery begins, to the extent permitted by law. Orders are therefore non-refundable after download or access is provided. Contact support@ai-datasets.uk promptly if a file is corrupted, inaccessible, or materially different from its description; we will investigate and, where appropriate, provide a replacement, corrected file, or other suitable remedy. Nothing in this policy limits mandatory consumer rights that cannot lawfully be excluded.