Hybrid feature selection framework for enhanced credit card fraud detection using machine learning models.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Electronic payments are widespread, but rising credit card fraud leads to significant financial losses.
• Detecting credit card fraud is difficult due to highly imbalanced datasets where fraudulent transactions are rare.
• Existing methods struggle with the inherent data imbalance, necessitating improved feature selection techniques.

Purpose of the Study:

• To propose a novel hybrid feature selection framework to enhance machine learning-based credit card fraud detection.
• To integrate Pearson correlation, information gain (IG), and random forest importance (RFI) for optimized feature selection.
• To validate the framework's effectiveness on diverse datasets and with various machine learning algorithms.

Main Methods:

• A hybrid feature selection framework combining Pearson correlation, information gain (IG), and random forest importance (RFI).
• Pearson correlation removes redundant features, while IG and RFI assess feature relevance.
• A union operation merges selected features for comprehensive and efficient selection, tested on PCA-transformed and real-world datasets.

Main Results:

• The proposed hybrid feature selection framework significantly outperformed baseline approaches across five diverse datasets.
• Superior fraud detection performance was achieved using machine learning algorithms like Random Forest, XGBoost, and CatBoost.
• The methodology demonstrated robustness and adaptability in enhancing fraud detection capabilities.

Conclusions:

• The novel hybrid feature selection framework offers a practical and effective solution for credit card fraud detection.
• The approach addresses the challenge of imbalanced datasets, improving the accuracy of machine learning models.
• This framework has the potential to serve as a real-time decision support system, benefiting the financial industry.