Anomaly Based Intrusion Detection System Using Soft Computing and Data Mining Approach

Abstract

Data and application security is most essential in today environment due to highly resource utilization in network environment. Various network attacks detection and prevention techniques has already introduced by various researchers in many existing systems. Two identification of malicious behaviour from large traffic and take action against such request is the part of IDS. Various machine learning techniques also already developed to generate strong rules with different Optimization algorithms. But still IDS facing some issues like unknown attack detection accuracy, low accuracy for network attacks etc. However, cyber security threats are also growing as the contact points to the Internet are increasing. A significant security issue today is the intrusion detection system (IDS). A Network Intrusion Detection System (NIDS) helps system administrators to detect violations of network security within their operations. However, many problems arise when a robust and efficient NIDS is developed for unexpected and unforeseeable attacks. In this work, a deep learning based approach implement for effective and flexible NIDS. It is confirmed that the deep neural network is effective for NIDS through the performance test. System uses Recurrent Neural Network (RNN) which is supervised learning algorithm to detect known and unknown attacks into the both environments. Initially, Data pre-processing has done with Weka tool and define standard technique to eliminate unwanted records for attribute values. The proposed RNN algorithm works in both models for training and testing respectively. In first section we train the model with different network intrusion data sets (KDD CUP99, NSLKDD, ISCX, NB-15 etc.). Once a rule has created system deals with testing model an imbalance data generation environment. The partial implementation introduce proposed RNN provides better accuracy then other machine learning techniques. Additionally, we are evaluating and comparing different deep learning algorithms, namely RNN, CNN, DNN and PNN algorithm on cloud environment to detect intrusion in the network.