Intrusion Detection System Using Auto-Encoders

Abstract

In today's increasingly interconnected society, network assaults are serious concerns and so is its security. To deal with these security threats various methods have been implemented both in network and host system. These procedures are broadly categorized as Intrusion Detection System (IDS). Recent studies have applied traditional machine learning to detect network attacks by studying the patterns of network behaviors and training a model for classification. These models typically include large labeled datasets, but the rapid speed and unpredictability of cyber-attacks make it difficult to mark them in real time. In order to identify the attacks in real time, any intrusion detection system has to deal with unknown attacks is more complex than previously defined attack. In this paper, Ensemble Autoencoders Based Intrusion Detection is proposed to deal with unknown attacks. Ensemble based model is already known for its increased accuracy and the  smaller, less complicated Autoencoders would also reduce the system processing power overhead. Variation auto-encoders are being used in our ensemble model. After getting the output from the ensemble models the output is fed into a classifier to detect intrusion and normal data.