COMPARATIVE ANALYSIS OF VARIOUS ALGORITHMS WITH DEEP NEURAL NETWORK FOR INTRUSION DETECTION SYSTEM

Abstract

An intrusion detection system (IDS) uses a lot of machine learning techniques to find and classify cyberattacks at the organization and host levels in a timely and independent way. However, a scalable solution is required because aggressive assaults are constantly evolving and occur in such large numbers. The network protection local area can secretly get to different malware datasets for additional investigation. Be that as it may, no ongoing review has analyzed the presentation of different AI calculations utilizing an assortment of secretly open datasets from top to bottom. Due to the dynamic nature of malware and its constantly shifting attack strategies, the privately provided malware datasets must be properly optimized and benchmarked. This investigation focuses on a deep neural network (DNN), also known as a deep knowledge model, to develop an adaptable and effective intrusion detection system (IDS) for describing and classifying shifting and unexpected cyberattacks. It is anticipated to evaluate various datasets created over time using static and dynamic procedures due to the constant change in network structure and the rapid definition of attacks. This kind of research can connect to the swish algorithm, which can accurately predict threats in the future. On several privately accessible standard malware datasets, a comprehensive evaluation of trials of DNNs and other conventional machine learning classifiers is presented. The following hyperparameter selection methods are used to select the ideal network parameters and network topologies for DNNs using the KDDCup 99 dataset. Over the course of one thousand DNN experiments, the knowledge rate fluctuates between 0.01 and 0.5. To comply with the standard, the DNN model that performed well on KDDCup 99 is applied to a variety of datasets, including NSL-KDD, UNSW-NB15, Kyoto, WSN-DS, and CICIDS 2017. Through several protected layers, our DNN model acquires the abstract and high-dimensional point representation of the IDS data. DNNs perform better in standing out from ordinary ML classifiers, according to exhaustive exploratory testing. Finally, we provide scale-crossbred-IDS-Alert Net, a cold-thoroughbred DNNs framework that is significantly scalable, can be used in real time to cover network business and host-location events effectively, and is able to proactively notify potential cyberattacks.