AN APPROACH TO DEVELOP THE HYBRID ALGORITHM BASED ON SUPPORT VECTOR MACHINE AND NAÏVE BAYES FOR ANOMALY DETECTION

Abstract

Anomaly detection is an important problem that has been researched within diverse research areas and application domains. Many anomaly detection techniques have been specifically developed, and the existing research have low detection rate. This research work proposes a weighted sum formulation for ensemble of Support Vector Machine and Naïve Bayes for anomaly detection, k-fold cross validation to evaluate the error metric associated with a candidate ensemble model and accuracy based weighting scheme to determine the weight values for member algorithms. The experiment has been conducted in 10% (Knowledge Discovery and Data Mining) KDD dataset. The data has been preprocessed to remove the duplicate records. The categorical data in the 10% KDD dataset has been converted to numeric value using binary encoding scheme. The features of the dataset have been selected using information gain. The grid search has been applied to the dataset using 10-fold cross validation to determine the parameters for Support Vector Machine (SVM). The SVM has been implemented using RBF kernel and value of gamma and C of 0.0001 and 1 respectively. The hybrid algorithm has been implemented to combine the outcome of prediction of SVM and Naïve Bayes classifiers using weight factors. The weights factors have been calculated using root mean square error of prediction as error metric. The classifier with high accuracy has been given higher weight and classifier with the lower accuracy has been given lower weight. For the validation of result, ten-fold cross validation has been employed. The performance of SVM classifier, Naïve Bayes classifiers and hybrid algorithm has been compared using Receiver Operating Characteristic (ROC) curve and classification metrics.