votrananhquan/fraud-detection-model

Credit Card Fraud Detection — Champion Model

Model Description

Champion model for credit card fraud detection on the Kaggle Credit Card Fraud Dataset. Registered in MLflow Registry as fraud-detection-model@production.

Trained under a leak-free protocol: stratified 64/16/20 train/validation/test split, the Amount scaler fitted on the training split only, early stopping watched on validation, and the champion selected by validation PR-AUC. The test split influences no decision and is scored once for reporting.

Performance

Selected on validation (n/a rows):

Metric	Value
PR-AUC	0.7407
Recall	0.8354
Precision	0.5238
F1	0.6439

Reported on the held-out test split (n/a rows):

Metric	Value
PR-AUC	0.7462
Recall	0.8878
Precision	0.4555
F1	0.6021
True positives	87
False positives	104
False negatives	11

Usage

import joblib
import numpy as np

model = joblib.load("baseline_lr.pkl")
# features: V1-V28 (PCA, pass through), Amount scaled with the TRAINING-split
# StandardScaler: mu=87.9702, sigma=245.5762
X = np.array([[...]])  # shape (1, 29)
proba = model.predict_proba(X)[:, 1]  # fraud probability

MLflow Tracking

• Run ID: 176f185e3b7a400dba35d6f841de31e2
• Model version: 3
• Registered name: fraud-detection-model