Security-Trust-Determinism Co-Design Using Hybrid Intrusion Detection With Temporal Modeling for Real-Time Publish-Subscribe Middleware

This paper proposes a trust-aware hybrid intrusion detection (IDS) for real-time pub/sub middleware that jointly targets security effectiveness and determinism preservation. The proposed approach fuses a low-latency inline detector with a temporal convolutional network (TCN) through a calibrated score-level combination, and then aggregates the fused evidence using a lightweight Bayesian trust engine to maintain continuous trust scores for middleware entities. Trust-guided enforcement is constrained by deterministic quality of service (QoS) budgets to avoid degrading real-time guarantees. Experiments on the publicly available UNSW-NB15 benchmark demonstrate that the hybrid design improves separability and robustness to time-dependent attack behaviors while remaining compatible with real-time execution. The obtained results of the proposed work establish a practical security–trust–determinism trade-off for deploying hybrid machine learning-based IDS in real-time middleware environments.