Comparing Machine Learning Algorithms and DNN for Anomaly Detection

Abstract

Cyber space became inevitable in today's world. It needs a security technology to safeguard the whole system from outsiders. An intrusion detection system acts as a strong barrier and screens the vulnerability. There is an upgraded amount of network attacks such as DoS (denial of service), R2L (remote to local) attack, U2R (user to root), and probe attack. These network attacks lead to prohibited usage of data from various applications like medical, bank, car maintenance, and achieve activities. This will result in financial gain and prevent authorized persons from accessing the network. Intrusion detection systems were implemented in systems where security is desirable. The conventional system makes use of machine learning techniques such as random forest and decision trees that entail many computational resources and higher time complexity. To overcome this, a DNN-based intrusion detection system is proposed. This IDS not only detects the abnormalities but also results in higher accuracy compared to existing systems. This also improves the speed, accuracy, and stability of the system.