Optimal attack detection using an enhanced machine learning algorithm

As computer network and internet technologies advance more quickly today, the importance of network security is widely acknowledged. This research intends to introduce a new security platform for SaaS framework, which comprises two major phases: (1) Optimal Feature Selection and (2) Classification. Initially, the optimal features are selected from the data set. A novel algorithm named Accelerator updated Rider Optimisation Algorithm (AR-ROA), a modified form of ROA and Deep Belief Network (DBN) based Attack Detection System is proposed. The optimal features that are selected form AR-ROA are subjected to DBN classification process, in which the presence of attacks is determined. The proposed model outperforms other traditional models in aspects of Accuracy (95.3%), Specificity (98%), Sensitivity (86%), Precision (92%), Negative predictive value (97%), F1-score (86%), False positive ratio (2%), False negative ratio (10%), False detection ratio (10%), and Matthew's correlation coefficient (0.82%).