A Hybrid Approach for Intrusion Detection in the Internet of Things Using Harris Hawks Optimization and Deep Learning Algorithms

Intrusion detection in Internet of Things (IoT)-based smart cities is essential due to the increasing volume and complexity of cyberattacks. Traditional detection systems face two major challenges: achieving high accuracy and minimizing false alarms, particularly in large-scale and heterogeneous IoT networks. This paper proposes a novel hybrid intrusion detection system that combines Harris Hawks Optimization (HHO) for feature selection with a multi-layer neural network enhanced by learning automata for adaptive classification. The HHO algorithm efficiently reduces input dimensionality by selecting the most relevant features, while the learning automata optimize the network's weights dynamically, improving training stability and robustness. The proposed system is evaluated using the KDDCup99 dataset under both binary and multiclass scenarios. Experimental results show an average accuracy of 96.53%, a true positive rate (TPR) of 94.91%, and a false positive rate (FPR) of 2.80%. Compared to recent baseline models, the proposed method demonstrates superior performance in accuracy and false alarm reduction, confirming its suitability for real-time intrusion detection in dynamic IoT-based environments.