Hybrid Ensemble-CNN with Rejection Mechanism for Cybersecurity Support

Abstract

In cyber security applications where the consequences of mis-prediction can be very serious, these Reliable classification systems are critical. This paper presents a hybrid classification method based on ensemble learning models and CNN architectures and a confidence-based rejection method that enhances the predictive confidence. The Iris dataset is used as a benchmark dataset to test the classification behavior of the framework prior to its deployment in a cyber security driven decision-making scenario. A light-weight 1D CNN architecture is integrated with traditional ensemble classifiers such as Random Forest, Gradient Boosting and AdaBoost. An rejection mechanism is added to prevent low confidence predictions in uncertain scenarios and thus enhance trustworthiness. Experimental results provide a proof of the superiority of the proposed hybrid framework in terms of classification accuracy, precision, recall and F1-score, when compared to the individual models, in terms of risk of misclassification in the case of selective rejection. The results illustrate the applicability of the framework in the cybersecurity field, such as intrusion detection, malware classification, and anomaly detection, areas in which decision making is critical and must be automated.