Ensemble probability meta-classifier for attack detection leveraging user behavior and big data analytics

Abstract

The rapid rise in cyber threats requires effective detection methodologies to accurately identify malicious behavior. Conventional detection techniques often struggle with the volume and complexity of data generated in today’s computing environments. This study focuses on detecting attacks by analyzing user behavior and describes an ensemble-based approach. The proposed EPMC algorithm integrates multiple base classifiers, including transformer models, deep neural networks (DNN), and gated recurrent units (GRU) to enhance detection accuracy and robustness. The EPMC method combines the probabilistic outputs of these classifiers and integrates them through a softmax-based fusion mechanism, which enables data-dependent weighting of base classifiers. The fused representation is then used to train a meta-classifier for final prediction. Extensive studies with large-scale datasets show that our technique is excellent for capturing unusual user activity. Overall, these results suggest that our EPMC-based model is able to achieve accuracy of 95%, recall of 96% and F1-score of 95%. The technology proposed in this paper, integrated with big data analytics offers a effective detection of intricate attacks and adapts to the changing safety conditions. These results tend to improve cyber defense capabilities and show the uses of ensemble learning techniques for deployment in existing security solutions.