Intrusion Detection System by Outlier Detection methods using Deep Learning on Cloud

Abstract

In the recent time, the growth in the internet-based service management have increased to a greater extend. The traditional locally managed services for academia, research and for industry have moved into internet-based applications. Thus, the application and the data are now accessible by the consumers over the internet. This paradigm has brought the feasibility to the customer to perform functional task related to business operations, where the application and the data generated by the application can be highly critical to be exposed. Hence, the developers of these applications imply multiple security restrictions on the application as well as on the data. Nevertheless, the hackers find their way to intrude the system and damage the critical information. Hence, the primary challenge of the present research is to detect and in possible situations terminate the intrusions from the system. Multiple research attempts have aimed to detect the intrusions using many methods, however due to the continuous attempts and sophisticated algorithms deployed by the intruder, none of the attempts could reach to their best performance or get outdated in very less time. Henceforth, this work aims to design a novel approach for detection of the intrusions in the system. This work after a detailed analysis have identified that the analysis of characteristics for the applications deployed with the primary system can help in detecting the intrusion. However, the other applications with the primary application can be a support feature to the main applicationand often perform some tasks which can appear as intrusion. Hence, detection of the accurate application and labelling the application as intrusion is one of the prime challenges. Thus, this work deploys a deep machine learning-based outlier detection method to identify the intrusions. The work demonstrates nearly 93% accuracy in detection of the intrusion and in parallel also defines a novel method for deep outlier detection with nearly 70% accuracy and for 100% accuracy for removal of the outliers.