Automated network intrusion detection using multimodal networks

Intrusion detection requires accurate and timely detection of any bad connection that intends to exploit network vulnerabilities. Previous approaches have focused on deriving statistical features based on domain knowledge, followed by primitive machine learning and ensemble techniques. Grouping all the parameters as a single input to a model may not always be effective. In this paper, we propose using multimodal networks for network intrusion detection. The input logs are segregated into multiple sub-groups trained differently. Their intermediate representations are combined to produce the final prediction. This approach handles the strengths of individual features better as compared to normalisation. The system is evaluated on the NSL-KDD dataset and is compared with standard methods across multiple performance metrics. The proposed system achieves an accuracy of 83.5, highest as compared to other approaches. Channelling inputs for richer feature extraction is fast gaining traction, and we extend the same in cybersecurity.