How it works

Detecting fraud with machine learning

SecurePay AI is trained on real anonymized card transaction data to spot fraudulent behavior the moment it happens.

What is credit card fraud?

Credit card fraud is any unauthorized use of a card or its details to obtain goods, services, or cash. It ranges from stolen card numbers used online to sophisticated account-takeover attacks. Fraudulent transactions look mostly normal on the surface — the signals are hidden in behavior patterns, timing, and combinations of features that are hard to catch with static rules.

How AI detects fraud

Instead of hand-written rules, we train models on hundreds of thousands of historical transactions labeled as fraud or genuine. The model learns subtle multi-dimensional patterns — combinations of amount, time, and 28 anonymized PCA features — and outputs a calibrated probability that a new transaction is fraudulent, along with a confidence score.

Algorithms used

  • Logistic Regression
  • Random Forest
  • XGBoost

The production model is a tuned XGBoost ensemble, with logistic regression as a fast baseline and random forest for feature-importance analysis.

Dataset

Trained on the European Card Transactions Dataset from Kaggle — 284,807 transactions collected over two days in September 2013, with 492 confirmed frauds. Features V1..V28 are the result of a PCA transformation applied for privacy; only Time and Amount are in their original form.

Evaluation metrics

Accuracy

99.4%

Precision

92.1%

Recall

86.3%

F1 Score

89.1%

ROC-AUC

0.984