Improving Cloud Security with Intrusion Detection based on Classifier Combination

Research Aim: By development of information technology, network security is considered as one of the main issues and has great challenges. Intrusion detection systems are a major component of a secure network. Traditional intrusion detection systems cannot adapt themselves to the new attacks thus today's intrusion detection systems have been introduced based on data mining. Research method: Identifying patterns in large volumes of data, is a great help to us. Data mining techniques by identifying a binary label (normal packet, abnormal packet) and specifying attributes by classification algorithms can recognize the abnormal data. Therefore, the precision and accuracy of intrusion detection systems will increase, there by network security increases. Findings: In this paper, we propose a model that evaluates the performance of the different algorithms on dataset. Simulation results show that the decision tree of the J48 algorithm, neural network of the Neural net algorithm, Bayesian network of the HNB algorithm, laze model of the K-STAR algorithm, LibSVM in support vector machine algorithm, and Rule Induction Single Attribute algorithm in the rule-based model, have the best result in the different parameters for performance evaluation of intrusion detection system. J48 algorithm provides the highest performance in the all-mentioned algorithms which has the accuracy of 85.49%, the precision of 86.57% and the recall of 86.90% for intrusion detection system. Conclusion: The main innovation in this paper is using the laze model algorithms that are not used in the intrusion detection systems. Also, we propose the 5 different samples from primary extracted data that achieve the best results for the different models and algorithms.