EKNIS: Ensemble of KNN, Naïve Bayes Kernel and ID3 for Efficient Botnet Classification Using Stacking

Any efficient Bot-Detection tool must be able to classify bot activity as, `bot' with utmost accuracy. The key factor that influences the efficiency of a Bot-Detection tool is the selection of a classification algorithm whose prediction accuracy is the maximum. This paper proposes the implementation of a Hybrid Approach involving k Nearest Neighbor (kNN), Naïve Bayes and ID3 classifiers resulting in most encouraging prediction accuracy values. The proposed scheme is followed after the preprocessing phase that involves the computation of weights of each attribute in the data set through different Weight Based Feature Selection algorithms and determining the mean of weights of each attribute. A final Mean of mean values of all the attributes is computed. All those attributes whose mean values fall below the final Mean of mean values are discarded while all other remaining attributes are considered significant attributes and are used for botnet classification.