An Efficient Intrusion Detection Model Employing Improved Boruta Search Optimization Approach for IoT Applications

The rising frequency of cyberattacks has made Network Intrusion Detection Systems (NIDS) essential for securing IoT environments. However, NIDS performance largely depends on the learning mechanism and quality of training data. Traditional rule‐based methods struggle to process the massive volume of network traffic generated within seconds, prompting the adoption of data mining techniques to enhance detection accuracy. Effective feature selection is crucial for improving detection rates and computational efficiency but still faces challenges like high dimensionality, false positives, and reduced accuracy. To overcome these limitations, this study introduces a modified Boruta Search and Krill Herd Optimization‐based (BS‐KHA) classification model for precise intrusion detection across multiple IDS datasets. The proposed model achieved binary classification accuracies of 99.85%, 99.96%, and 99.92%, and multiclass accuracies of 99.93%, 99.99%, and 99.87% on NSL‐KDD, UNSW‐NB15, and CICIDS2017 datasets, respectively—surpassing existing IDS frameworks.