Credit Card Fraud Detection Using Machine Learning

Abstract: Credit card fraud is a rapidly growing threat fuelled by the expansion of e-commerce and online payment platforms. This paper presents an unsupervised machine learning system for detecting fraudulent credit card transactions using K-Means Clustering. The pipeline loads the Kaggle credit card dataset (284,807 transactions; 492 fraud, 0.172%), drops the ‘Time’ column to simplify clustering and reduce temporal bias, applies StandardScaler normalization, and trains K-Means with K=10 clusters. Fraud is detected by identifying the cluster with the highest fraud density using idxmax(). Model quality is measured by silhouette score, and classification performance is evaluated via confusion matrix, precision, recall, F1-score, and accuracy. Due to the extreme class imbalance, precision, recall, and F1-score are the primary evaluation metrics rather than accuracy. PCA (n_components=2) is used to produce a 2D cluster visualization. A Gradio Blocks interface is deployed with a ‘Single Transaction’ tab (30-feature input, Generate Random button) and a ‘Batch Prediction’ tab (CSV upload). All metrics reported are computed directly from running the project code.

Keywords: credit card fraud detection, K-Means clustering, unsupervised learning, StandardScaler, PCA, silhouette score, Gradio, confusion matrix, fraud density