An Ensemble-based Outlier Detection Approach on Intrusion Detection

Outlier detection is becoming an utmost essential for data pre-processing and modeling. Commonly, the model cannot minimize the training errors during the learning process due to outliers in the training dataset. Several outlier detection approaches have been proposed to address a variety of problems in different disciplines. In this study, an ensemble-based outlier detection approach is proposed to detect the outliers on intrusion datasets. The proposed ensemble approach consists of a local outlier factor, isolation forest, and one-class SVM method to identify the outliers on intrusion datasets. The majority voting method is used to finalize the connection as an outlier by averaging the outlier indexes generated by the approach. The result of the proposed method is compared before and after detection of the outlier obtained from the popular supervised classification methods. Empirically, we have achieved a better classification result and drastically reduced the training error by eliminating the outliers during the learning process.