A Novel Hybrid Intrusion Detection Framework for IoT Networks Using Mother-Infant Communication Metaheuristic and Attentionbased Deep Learning

The rapid proliferation of Internet of Things (IoT) devices has exponentially expanded the cyber-attack surface, rendering conventional intrusion detection systems (IDS) increasingly ineffective against sophisticated zero-day threats. This paper presents a novel hybrid IDS framework that synergistically integrates the Mother-Infant Communication (MIC) metaheuristic algorithm with an attention-based deep learning architecture. The MIC algorithm, inspired by echolocation behaviors in animal colonies, is employed for the first time in cybersecurity to perform dual-phase optimization: optimal feature selection and hyperparameter tuning of a deep classifier. The core detection engine is a hybrid Convolutional Neural Network-Bidirectional Gated Recurrent Unit (CNN-BiGRU) augmented with a hierarchical attention mechanism, mathematically formulated to capture both spatial and temporal dependencies in network traffic. The framework is rigorously evaluated on the CIC-IoT-۲۰۲۳ and ToN-IoT datasets. Experimental results demonstrate that MIC-driven optimization achieves significant performance enhancement, improving detection accuracy by ۵.۲% to ۶.۸% compared to baseline models without optimization, achieving peak accuracy of ۹۹.۶۳%. The mathematical analysis reveals that the bidirectional search mechanism of MIC provides superior convergence properties in the optimization landscape of high-dimensional feature spaces.

کلیدواژه ها:

Intrusion Detection System ، Internet of Things ، metaheuristic optimization ، Deep Learning ، attention mechanism