Dimensionality Reduction for Network Intrusion Classification Based on Dolphin Mating Algorithm

The rapid rise of the internet and other associated technologies, along with other things, has caused a huge increase in the sharing of sensitive information across multiple networks. Because of this, fraudsters are getting better at their attacks, which shows how important it is to have strong cyber security. Any network security plan needs to have intrusion detection systems (IDSs) as a key feature. IDSs let experts and administrators keep an eye on possible threats by watching network data for signals of suspicious or hazardous activities. Anomaly-based systems employ machine learning (ML) to find suspicious activity in network traffic. This work suggests a feature selection (FS) structure for building IDSs based on anomalies by combining heuristic and ML methods. This proposed study aims to enhance IDS performance regarding attack detection and detection time by employing heuristics to select attributes. In the proposed approach, Dolphin Mating Algorithm (DMA) will be employed to implement the FS mechanism and a customized convolutional neural network (CNN) deep learning model will be implemented for the classification job. The adopted dataset is the NSL-KDD. Out of 41-features that consisting the NSL-KDD dataset, only 9-attributes were selected, in other words, 21.95% of the features will be used to train the CNN model. More than 90% accuracy for the four classes of the attack types was achieved.