Published Paper Details:

Within the previous decades, there has been increase in the technology which led to the usage of many devices thereby causing threat to personal data, corporate data, device and network itself. This gave rise to the advancements in Intrusion Detection Systems (IDS) which act as defense mechanism. The Intrusion Detection Systems helps to detect unintended access to network and device with the use of several strategies to identify threats in the network. The existing systems require human interaction to analyze and to identify the threats. However it's a major drawback as it is difficult to human to notice the intrusions from various sources of network traffic. Intrusion Detection System can automatically check for network intrusions without the use of human to interact with the systems. The main idea behind as IDS is to detect the threats based on analyzing and predicting the behaviors of users, these actions are going to be used by our system to train our IDS to enhance the detection of upcoming threats well in advance or at the time of attack happening. No existing system is not very much sure about the attacks and is prone to attacks and may fail to accurately detect the attack. The advancements in technology also led to the changing behavior of attacks and these attacks are to be monitored continuously to fine tune our systems for best results. A Deep Neural Network (DNN), which works in multilayered architecture, is used for the development of IDS for efficiency and accuracy to notify unknown and unforeseen attacks performed on the network. The continual updating and analysis of network behavior is required for the system to classify attacks within a short span of time. For this experiment we worked on preexisting dataset like NSL-KDD, KDDCUP1999 for training our system and to test the working of our system. The Dataset has 41 different types of attack these attacks are classified and grouped into 4 categories like Denial of Service (DOS), Unauthorized access to root Privileges (U2R), Unauthorized access from Remote System (R2L), Probing. Based on the test results the proposed system model works better than previously existing systems the training time is lesser than that of existing once and loses are reduced with 96% accuracy test results, which is 3% more than the existing study.