Abstract:
“The Internet of Things (IoT) is becoming more and more prevalent, which gives cybercriminals a larger assault surface. The effectiveness of machine learning for intrusion detection in Internet of Things security is examined in this work. A dataset from the ML Repository is used in the study to examine the effectiveness of Decision Tree (DT), Random Forest (RF), KNearest Neighbors (KNN), XGBoost, AdaBoost, and Logistic Regression (LR). RF, DT, and XGBoost each scored a remarkable 0.99 accuracy, while KNN came in at 0.94.A stacking ensemble strategy is further investigated in the paper, combining LR as the meta learner with RF, DT, XGBoost, AdaBoost, and KNN as base learners. Accuracy scores for this ensemble were 0.99 for RF, DT, and XGBoost, 0.93 for KNN, and 0.98 for the meta learner LR. Overall performance was enhanced. Notably, because the dataset was balanced, SMOTE was not found to be necessary in order to correct class imbalance. According to our research, machine learning, intrusion detection, and IoT security are three ensemble approaches that are highly effective at protecting IoT environments. By demonstrating the usefulness of ensembles for obtaining reliable intrusion detection, this study advances the practice. Stacking ensemble is a possible approach that may help mitigate overfitting even while individual models performed well. For even more dependable IoT security systems, our work opens the door to future research into the best model selection and hyperparameter tuning.
