Implementation of shallow neural network to intrusion detection system

Abstract

The development of increasingly widespread computer networks makes users pay more attention to data security and convenience in accessing it. In addition, heavy data traffic on the network and threats and interference, make users feel worried about exchanging data. In anticipating this, users need a system capable of detecting intrusions on their computer network. This research proposes an intrusion detection system that can be the leading solution for network users. The development of this system certainly requires a reliable detection algorithm for classifying the types of data traffic. Therefore, machine learning techniques, which are very popular in Artificial Intelligence in processing large-scale data, are applied in this study. The development of this Intrusion Detection System adopts CRISP-DM by utilizing a Shallow Neural Network to build a classification model. Furthermore, this method will learn to use intrusion data prepared to produce a Neural Network model capable of detecting attacks based on existing data traffic. This model was developed based on a new dataset called ALLFLOWMETER HIKARI2021. This data is divided into 70% training and 30% testing data. Based on ten modeling experiments, the resulting model is capable of producing an MSE and an accuracy of around 0.04438 and 95.562% at the testing stage so that the developed system can detect various network intrusions that exist during the testing phase, especially in distinguishing between two types of data traffic, namely benign or malicious.