Improving performance of classification intrusion detection model by Weighted extreme learning using behavior analysis of the attack

This research was aimed to develop classification intrusion detection model by Weighted ELM which presented in [8], bringing analysis of 42 attributes to find the ones related to each format of attack, remaining only 13 attributes which were chosen to use in Weighted ELM working system in order to classify various attack formats and compared to experimental result with SVM+GA [7] and Weighted ELM techniques [8]. The result showed that New Weighted ELM was quite accurate in classifying every format of attack, which the presented working system of the method used RBF Kernel Activation Function and defined Trade-off Constant C value at 2² = 4, giving validity value to be Normal = 99.21%, DoS = 99.97%, U2R = 99.59%, R2L - 99.04% and Probing Attack = 99.13%, average validity value was at 99.39% Comparing to Weighted ELM in [8], found that, the presented method could improve the effectiveness of the former method enable to more classify R2L from 93.94% to 99.04%, and from 96.94% to 99.13% for Probing Attack meanwhile DoS and U2R had lower effectiveness, yet there was resemble effectiveness.