Network Packet Breach Detection Using Cognitive Techniques

Abstract

Machine learning approach is being extensively used in the area of cybersecurity in recent years developing solutions to protect Internet users. The use of state-based cognitive data and the increased prevalence of data mining has allowed for the amalgamation of statistical concepts with machine learning providing real-time network packet analysis with an aim to detect when an entity has intruded the network. In this paper, the use of mean squares error for packet payload aggregation, coupled with prediction techniques using Bayes and ensemble learning outputs to data clusters provide useful and important insight to generate hybrid solutions to existing data breach problems. The use of dynamic tolerance levels and countering this against the potential for false positives is central to the design of our proposed scheme. We believe that correlations between expected information against the aggregated payloads could provide sufficient level of accuracy, which is sufficient to flag certain packets for further human assessment.