<![CDATA[Background: The proliferation of IoT devices increases network exposure to sophisticated attacks such as DDoS, demanding robust intrusion detection. Specific Background: Traditional ML-based IDS face challenges with high-dimensional data and evolving attack patterns. Knowledge Gap: There is a need for automated feature selection that preserves detection performance while reducing complexity for large modern datasets. Aim: This study proposes an Enhanced Flower Pollination Algorithm (EFPA) for optimal feature selection combined with an ensemble classifier (Random Forest, ID3, SVM) to improve IoT intrusion detection. Methods: The model was evaluated on NSL-KDD and UNSW-NB15 with preprocessing, SMOTE balancing, and 70:30 train–test splits. Results: The EFPA-selected features with ensemble voting achieved 99.67% accuracy on UNSW-NB15 and 99.32% on NSL-KDD. Novelty: Integration of EFPA for dimensionality reduction with ensemble classification on modern benchmarks. Implications: The approach reduces computational load while maintaining high detection performance, suggesting promise for scalable IDS in IoT environments. Highlights: EFPA reduces feature set while preserving detection accuracy. Ensemble voting improves generalization across benchmarks. High accuracy achieved on UNSW-NB15 and NSL-KDD. Keywords: EFPA, Ensemble Classifier, Intrusion Detection, NSL-KDD, UNSW-NB15]]>
Copyrights © 2025
