Hybrid AI Model for Online Payment Fraud Detection Using Machine Learning (Iconic Research and Engineering Journals)

Summary

• The proposed hybrid AI system was trained and evaluated on a transaction dataset containing more than 6 million records with transaction time steps ranging from 0 to 743.p.6
• Individual machine learning models achieved baseline evaluation accuracies ranging between 99.87% and 99.93%.p.1
• On a full test set of 1,272,524 transactions, Random Forest identified 1,477 of 1,620 fraud cases with 354 false positives.p.19
• XGBoost achieved an overall accuracy of 99.88% on the full test set, detecting 1,591 out of 1,620 fraud cases alongside 1,464 false positives.p.19
• Evaluations on a 100,000-record test subset showed XGBoost reaching an accuracy of 99.928% and Random Forest achieving 99.911%.p.21
• The soft-voting classifier recorded a overall performance accuracy of 99.906% on the 100,000-record evaluation subset.p.21

Key insights

1. The hybrid framework integrates SVM, Random Forest, and XGBoost classifiers whose predictions are combined using a soft-voting ensemble.p.1
2. Class imbalance in financial transaction data is addressed by applying Synthetic Minority Over-sampling Technique (SMOTE) exclusively to the training set.p.1
3. Random Forest delivered a balanced performance on the full test set with a fraud precision of 0.81 and a recall of 0.91.p.19
4. XGBoost demonstrated high fraud recall at 0.98 on the full test set, but generated 1,464 false positives resulting in a fraud precision of 0.52.p.19
5. Evaluating on a 100,000-record subset revealed that SVM and Random Forest attained 1.00 fraud-class precision but suffered lower recall.p.21
6. The deployment application is designed around Flask REST API endpoints and a persistent MySQL database running on XAMPP.p.5
7. Feature scaling uses StandardScaler fitted strictly on training data and applied to both train and test sets to maintain data integrity.p.7