Hybrid Frameworks Combining Meta-Heuristics and Machine Learning Models for Feature Selection In Credit Card Fraud Detection

Detecting credit card fraud is one of the main challenges in data-driven financial systems due to the severe imbalance of classes, the complexity of behavioral patterns and the constant variability of fraudulent methods. Machine learning algorithms have a high ability to model transactional data and their performance is directly dependent on the quality of the feature space. The presence of redundant or under-informative features can reduce the sensitivity of models to the minority class. Feature selection is considered as a key component in the design of fraud detection systems. In this paper, Hybrid frameworks are presented for integrating classification algorithms including Decision tree, Support vector machine, Naive Bayes and K-nearest neighbor with meta-heuristic algorithms as feature selection. Methods such as Genetic Algorithm(GA), Particle Swarm Optimization(PSO), Improved PSO(IPSO), Ant Colony Optimization(ACO), Firefly Optimization(FO), Rock Hyrax Swarm Optimization(RHSO) and Differential Evolutionary(DE) have been used to extract the optimal subsets of the features. The results on the ULB dataset show that meta-heuristic feature selection remarkably outperforms the no-selection setting, increasing Recall from ۰.۳۷ to about ۰.۷۹ for SVM and from ۰.۸۴ to ۰.۹۵ for Decision Tree, while improving F۱-Score from roughly ۰.۵۳ to above ۰.۸۰ and from ۰.۸۳ to over ۰.۹۰, respectively, with RHSO and GA based methods achieving the most consistent gain.

کلیدواژه ها:

Fraud Detection ، Feature Selection ، Metaheuristic Optimization ، Machine learning