Strengthening IoT Network Security: Automated Intrusion Detection Using a Stacked Classifier

Abstract

The rapid growth of Internet of Things (IoT) networks has increased the exposure of interconnected devices to sophisticated cyberattacks, creating a critical need for automated and reliable network intrusion detection mechanisms. Conventional intrusion detection techniques often face limitations in handling high-dimensional network traffic, redundant attributes, and imbalanced attack classes, which can reduce detection effectiveness and increase false alarms. The UNSW-NB15 dataset, containing contemporary normal and malicious network traffic with flow-based network features, is utilized for intrusion classification. Preprocessing includes data cleaning, feature selection using Filter and Wrapper methods, Lasso-based Embedded selection, Random Forest with Recursive Feature Elimination (RFE), and Pearson Correlation Coefficient (PCC) analysis, followed by Synthetic Minority Over-sampling Technique (SMOTE) to mitigate class imbalance. Random Forest, Decision Tree, AdaBoost, Bernoulli Naive Bayes, K-Nearest Neighbors, and Logistic Regression are evaluated alongside a Stacking Classifier integrating Boosted Decision Trees, Random Forest-based Bagging, and LightGBM. Performance is assessed using accuracy, precision, recall, F1- score, and related classification measures. The Stacking Classifier achieves the highest detection performance, demonstrating improved intrusion identification and reduced false-alarm tendencies compared with individual classifiers. The proposed framework strengthens automated IoT network security through optimized feature representation, balanced learning, and ensemble-based intrusion detection.