Metaheuristic and Data Mining Algorithms-based Feature Selection Approach for Anomaly Detection

Abstract

In response to the evolving landscape of data security and the rising threat of anomalies in various domains, this study presents an innovative approach for anomaly detection in data streams and extends its application to financial network security. Our research initially focuses on identifying relevant correlation ratios for anomaly detection in the commercial data traffic of some companies listed during the 2014–2020 period. Utilising genetic algorithms (GA), gray wolf optimisation (GWO) for dimensionality reduction, we process 96 different financial streams. Subsequently, we employ the support vector machine (SVM) classifier to categorise companies into anomaly classes based on their financial data. The SVM classifier's initial performance, considering all financial ratios, falls short of our expectations. Our research thus provides a comprehensive approach to anomaly detection, showcasing the versatility and effectiveness of our methods in financial data stream analysis and network security. By combining the strengths of metaheuristic methods in data analysis and advanced deep learning techniques in network security, our work offers comprehensive solutions for anomaly detection in both domains. To enhance the classification accuracy and streamline the ratios, we introduce metaheuristic algorithms. GWO, in particular, stands out, with a fitness function of 0.2940 and an accuracy of 70.06% after 31 iterations. This algorithm successfully extracts 9 crucial financial ratios. These GWO-extracted ratios are then integrated into the SVM classifier, resulting in an anomaly detection model with outstanding accuracy, precision, error reduction, and efficiency, achieving 75.83%, 66.80%, 33.2%, and 80.3%, respectively.