A hybrid spiking convolutional neural framework with extreme learning machine for enhanced anomaly detection in network security.

Real-time cybersecurity systems continue to have a significant problem in detecting breaches in dynamic and highly unbalanced network streams. An online evolving spiking neural network (OeSNN-UAD) and an integrated extreme learning machine (ELM) as an analytical readout optimizer are combined with spiking convolutional layers for spatiotemporal feature extraction in this paper's unified hybrid SCNN-OeSNN-ELM framework. In order to provide discriminative temporal representations, Gaussian Receptive Fields (GRFs) encode streaming inputs into spike trains, which are then processed by spiking convolutional layers. These characteristics are then supplied into a developing OeSNN reservoir whose output connections are analytically optimized by the ELM, allowing for low-memory operation, online adaptation, and quick, non-iterative learning without the need for labeled input. The suggested framework routinely outperforms current spiking-based baselines (Vacuum Spiker and Hybrid SNN-IOMT), as shown by experiments on the Numenta Anomaly Benchmark (NAB) under a streaming protocol and robustness validation on CIC-IDS2017. The model's relative detection increases are usually between 3 and 10% on NAB and 3 and 6% on CIC-IDS2017, and it delivers moderate but statistically significant improvements across precision, recall, F1-score, balanced accuracy, and MCC. The framework also shows enhanced latency and energy efficiency due to its event-driven spiking computing and analytical ELM optimization.