The Novel K–Nearest Neighbor and Back Propagation Recursive Least Squares Algorithms for Intrusion Detection Systems

Dr. Sadeq Al–Hamouz; Prof.Dr. Reyadh Naoum; Omar Al–Sadoon

Volume : IV, Issue : III, March - 2015

The Novel K–Nearest Neighbor and Back Propagation Recursive Least Squares Algorithms for Intrusion Detection Systems

Dr. Sadeq Al Hamouz, Prof. Dr. Reyadh Naoum, Omar Al Sadoon

Abstract :

This research is focusing on the most important security aspect which is the intrusion detection systems (IDS). The system is designed to detect and classify intrusions to four basic attacks that every computer or network is most likely to be exposed to. The design is based on Novel K–Nearest Neighbor (K–NN) machine learning algorithm , and on multilayer neural network(MLP) which was trained using back propagation recursive least squares algorithm(BPRLS). Several stages are included within the proposed system, first stage includes data environment, which represents NSL–KDD99 dataset for intrusion detection, and the stage of data codification, and the stage of data preprocessing and categories classification stage using Novel K–Nearest Neighbor (K–NN) machine learning algorithm, and the Neural Network training stage using Back Propagation Recursive Least Squares (BPRLS) Algorithm. The proposed intrusion detection scheme performance was evaluated using two hidden layers with two different numbers of neurons each time. The results confirmed the effectiveness of the proposed scheme in terms of the accuracy, precision, TPR (recall), TNR (specificity), FPR( false positive rate), FNR( false negative rate) and MSE (mean square error). Back propagation recursive least squares algorithm (BPRLS) recorded a detection rate of 0.959726 with false negative of 0.0402742 and mean square error of 0.2213 during experiment. The proposed intrusion detection system �results� was compared to other intrusion detection systems �results� which were designed using supervised learning such as enhanced resilient back propagation or unsupervised learning such as Kohenon.