Intelligent Intrusion Detection in Military IoT Networks Using Recursive Feature Elimination with Extreme Gradient Boosting

Intrusion detection is critical for securing Military IoT (MIoT) environments, where precision, reliability and rapid response are crucial to safeguarding sensitive operations. This research evaluates multiple Machine Learning (ML) models including Naive Bayes (NB), Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Trees (DT), Support Vector Machine (SVM), Bagging, Random Forest (RF), Gradient Boosting (GBoost), Adaptive Boosting (AdaBoost), and Extreme GBoost (XGBoost) for multi-attack classification on the IDSIoT −2024 dataset. To enhance performance advanced feature selection techniques, PCA and RFE have been implemented to identify the top 50 most significant features. The proposed RFE-XGBoost model achieved outstanding results with an accuracy of 97.95% and an F1 score of 0.9497 outperforming all other tested models. This demonstrates its improved capability to accurately classify diverse attack types in complex IoT ecosystems. The study highlights the novelty of integrating recursive feature elimination with XGBoost to focus on impactful attributes, significantly improving detection performance. This approach offers an optimized, scalable and robust intrusion detection solution designed to the high security demands of military IoT environments.