Attack Detection using DL based Feature Selection with Improved Convolutional Neural Network

Abstract

Decentralized wireless networks that may connect without a central hub are named Mobile Ad-hoc Networks (MANET). Attacks and threats of the most common kind can easily penetrate MANETs. Malware, APTs, and Distributed Denial of Service (DDoS) assaults all work together to make Internet services less reliable and less secure. Existing methods have been created to counter these assaults, but they either need more hardware, result in significant delivery delays, or fall short in other key areas like as energy consumption. This research therefore provides an intelligent agent system that can automatically choose and classify features to identify DDoS assaults. In this study, we provide an automated attack detector for MANETs based on a multilayer, (1D) convolutional neural network (CNN). Grey relational analysis classifiers are employed to screen attack levels in the classification layer because of their simple mathematical operation. The sunflower optimization technique is also used to fine-tune the classifier's weight. The research suggested a supervised feature classifier and fed the compressed data from an unsupervised auto encoder to it. In our experiment, conducted on the custom-generated dataset CICDDoS2018, the system outperformed state-of-the-art deep learning-based DDoS attack finding methods by a factor of 98%. Our suggested technique utilizes the freshest CICDDoS2018 dataset in combination with automated feature selection and classification to achieve state-of-the-art detection accuracy at a fraction of the processing time.