A Modified Cascaded Feed Froward Neural Network Distributed Denial of Service Attack Detection using Improved Regression based Machine Leaning Approach

The cloud and its applications have been growing rapidly in the last decade, but every coin has two sides: positive and negative. The good thing about that is that online work is growing quickly, but the bad thing is that attacks are becoming more and more powerful. In the current generation, cyber-attacks are one of the most threatening issues for cloud-based applications. Different cloud attacks are presented, in which Denial of Service (DDoS) is one of the perilous attacks that consume the network bandwidth as well as computing resources of a targeted system. Users of the cloud service now have protection against DDoS attacks that are caused by people who haven't been given permission. In order to prevent the D-DoS attack, this research work has presented a modified feature selection approach. The proposed machine learning approach is separated into three parts: training, testing, and validation. The training of the proposed method requires a previous attack dataset. In 2017, the Canadian Institute of Cyber Security presented a data set for D-DoS attacks. This attack contains four categories, which are labelled as “Slowloris,” “Slowhttptest,” “Hulk,” and “Begian.”. For weight optimization, the proposed method employs the Levenberg-Marquardt method, and for validation, a cascaded feed-forward neural network is used. It is well known that, the machine learning algorithms consume a huge amount of time to train models. Introducing a fast training for parallel processing model will make it an efficient tool to reduce the time consumed for training process. It has been shown that the presented approach gives an adequate improvement in accuracy, that is, 99.96%.