A Novel Feature Selection using Optimized Eliminated Iterative Distance Correlation for SDN-Enabled Traffic Anomaly Detection and Mitigation

Abstract

SDN-based traffic anomaly detection conventional techniques have been used to reduce the network traffic and their attacks. Sampled Density Peak technique has better detection performance based on clustering algorithm with the sampling adaptation and Unsupervised Cluster-based Feature Selection (UCFS) mechanism. However, the feature selection was not accurate due to the fixed threshold. Hence in this article, an optimized feature selection with Sampled DP technique is used. In this technique, a Backward Elimination Iterative Symmetric Uncertainty (BE-ISU) process is used to select the smallest subset of features. This removes the least influential features towards back which ensures sufficient classification accuracy. To tackle this issue, an Optimized Eliminated ISU (OE-ISU) process is proposed by using Fractional-Order Darwinian Particle Swarm Optimization (FODPSO) algorithm. Finally, the experimental result shows that the proposed technique achieves better performance with the Sampled-DP technique for reducing the data dimensionality.