Efficient Feature Selection for IoT Security: A Comparative Analysis of Swarm Optimization Algorithms in Attack Detection

As the Internet of Things continues to expand, securing this interconnected space becomes increasingly critical. This research presents a wrapper-based approach to feature selection, utilizing swarm optimization algorithms to enhance intrusion detection systems in the IoT paradigm. We implemented and compared several swarm optimization techniques—gray wolf optimization, cuckoo search optimization, firefly optimization, whale optimization, and ant lion optimization—integrated with decision trees, light gradient boosting machine, and Gaussian Naive Bayes classifiers. These combinations were used to select optimal feature subsets, with the accuracy of the models serving as the fitness measure. Upon identifying the best feature subsets, we constructed detection models and compared their performance against models utilizing the complete feature set. This comparative analysis was conducted on a large-scale dataset, with over 46 million data points, ensuring robust evaluation. The selected feature subsets, particularly those derived from gray wolf optimization and cuckoo search optimization, exhibited exceptional performance, achieving high accuracy and weighted F1 scores. The findings of this study show the effectiveness of feature swarm optimization techniques in feature selection toward the development of lightweight intrusion detection systems for resource-constrained environments like IoT. While showcasing the importance of feature reduction, the study also discussed the trade-off between feature reduction and the model's accuracy. This study highlighted the future research direction in applying advanced feature engineering techniques and ensemble techniques for developing intrusion detection systems for IoT.