Improvement of Existing Intrusion Detection Systems Through Neural Networks

Abstract

Recently, there is an increased rate of improved and unknown intrusions in the Intrusion detection system (IDS) field. In this study, artificial neural networks (ANN) model is proposed and integrated to develop the IDS execution on classification. Continual semi-supervised learning is the primary method conducted on artificial neural networks to create the model. Neural networks undergo pre-training with extensive benchmark datasets used for intrusion detection depending on how the user profiles have been created that have exact simulation of events and behaviors that are checked on the network. The evaluation domain is summarized from a unique set of features from the data set. The dataset are different due to the different user profiles. The ANN are trained for diverse attacks like Network infiltration, DDOS, Web attacks and many other arising attacks. The continual learning allows the model to learn latest advancements in deep learning and updates on recent anomalies detected. This improves the IDS to detect anomalies accurately. The main contribution of the proposed model to mitigate the gap identified is to assist in better classification of data, through the continual learning of neural networks, so that false positives that pass as normal data can be reduced in the system. It is demonstrated with experimental findings that the technique proposed can maintain a real time feedback to the anomaly.