Evaluation Metrics for a Hybrid Classification System Based on the Distributivity Equation and the UNSW-NB15 Cyberattack Dataset

The aim of the study was to apply and evaluate the usefulness of the hybrid classifier to detect network intrusion threats on a comprehensive dataset. Our approach is established on ensemble classifiers involving the distributivity law which aggregates classifiers accordingly. The paper includes results of experiments performed on the UNSW-NB 15 dataset i.e. a collection of network packets exchanged between hosts. A five-fold cross-validation technique was used for the performance evaluation of classifiers on the dataset using the Scikit-learn tool. We present the results of classification measures of the performance of our hybrid algorithm created from aggregations associated with the distributivity equation of selected classification algorithms (Multilayer Perceptron Network, k Nearest Neighbors and Naive Bayes) and compare them with the results of these individual algorithms on raw data.