New Strategy for Detecting Intrusion by Using c4.5 Algorithm

Intrusion Detection Systems (IDSs) based on sophisticated algorithms rather than signature-base detections provide a more effective way of building secure networks within the information society. Proposes a data-mining based technique to detect intrusions using binary classifiers with feature selection and multiboosting. Builds a model employing feature selection to create a more accurate binary classifier for each type of attack, improving attack detection. Approach aggregates binary classifier’s decisions and own decisions to assign a class for a given input. Potential bias of a binary classifier is alleviated by other decisions. Multiboosting reduces both variance and bias. Experiment results show that built approach is more accurate and cost-effective than existing methods. Considers intrinsic features, time-based features, host-based traffic features, and content-based features from records.