credit-card-fraud-detector

Credit Card Fraud Detector

LightGBM classifier that detects fraudulent credit card transactions (binary classification on the classic 284,807-transaction dataset).

Model Characteristics

• Type: Binary classification (LightGBM)
• Inputs: 29 PCA/anonymized transaction features (all except Time)
• Output: fraud probability in [0,1]
• Class imbalance: 0.17% fraud rate, handled via class weights

Data Overview

• Dataset: mlg-ulb/creditcardfraud (284,807 rows, 31 cols)
• Standard 80/20 stratified split

Evaluation Results

• Test AUC: 0.947
• Test Average Precision: 0.791
• Fraud recall @0.5 threshold: 76.5%

Files

• fraud_model.joblib — trained LightGBM classifier
• feature_names.npy — feature names (excludes Time/Class)

Usage

import joblib
model = joblib.load("fraud_model.joblib")

Writeup

A full case study is in the accompanying writeup (also published on Kaggle): Detecting Credit Card Fraud with Gradient Boosting: A LightGBM Case Study.

Provenance

Trained by Anuj Bolewar (anujbolewar on Kaggle) on the mlg-ulb/creditcardfraud dataset.