Comparative assessment of aggregated classification algorithms with the use to mining a cyber-attack dataset

Abstract

Currently, we observe an enormous growth in the frequency of using the Internet, which is also causing an increase in attacks on computer nets. These phenomena significantly raise the importance of the use of Intrusion Detection Systems (IDS). Classification systems are an essential part of a cyber-attack detection task by classifying the attacks based on certain criteria. The purpose of this research is to assess the relative performance of five extensions of well-known classification methods using the distributivity law. The results of this investigation can help in the design of classification systems that use several classification methods, namely k-Nearest Neighbor, Naive Bayes, Support Vector Machine, Random Forests, and Multilayer Perceptron Network can be employed to increase the accuracy of the classification. This method requires the use of some adequate aggregation operators (e.g. average functions and triangular norms/conorms) for which the distributivity law occurs. The work contains principally the results of experiments carried out on the KDD'Cup 99 dataset using WEKA (Waikato Environment for Knowledge Analysis) tool.