Tackling fraud detection with an enhanced Kepler optimization and ghost opposition-based learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Cybersecurity

Background:

• Conventional fraud and malware detection systems struggle with novel threats, class imbalance, and high-dimensional data.
• Increased online activity necessitates more robust and adaptive detection methodologies.

Purpose of the Study:

• To propose an advanced Fraud Detection (FD) methodology, BKOA-GOBL, enhancing the Binary Kepler Optimization Algorithm (BKOA) with Ghost Opposition-Based Learning (GOBL) for improved Feature Selection (FS).
• To address class imbalance using Random Under-Sampling (RUS).

Main Methods:

• The BKOA-GOBL integrates GOBL with BKOA to balance exploration and exploitation, prevent early convergence, and enhance search diversification.
• Random Under-Sampling (RUS) is employed to handle class imbalance in fraud datasets.
• Validation performed using k-Nearest Neighbors (K-NN) and XGBoost (Xgb-tree) classifiers on five real-world benchmarks.

Main Results:

• BKOA-GOBL achieved up to 99.96% classification accuracy and 81.82% feature reduction on several benchmarks.
• Consistently high precision, recall, ROC_AUC, and F1-scores demonstrated reliable detection, though some datasets presented challenges.
• Comparative analysis showed BKOA-GOBL's dominance over 12 Metaheuristic Algorithms (MHAs) and Machine Learning (ML) classifiers in accuracy and efficiency.

Conclusions:

• BKOA-GOBL is a robust, adaptable, and effective methodology for high-dimensional fraud and malware detection.
• The approach demonstrates statistical superiority and practical applicability in real-world scenarios.
• The integration of GOBL and RUS effectively addresses limitations of conventional detection systems.