Rapid growth in usage of Internet of Things (IoT) devices has created a situation where security is highly vulnerable, and people require more sophisticated and evolving solutions. Conventional security solutions cannot overcome the issue of heterogeneity, resource scarcity, and dynamism of IoT environments. This paper suggests the use of a machine learning-based Intrusion Detection System (IDS) to identify and attempt to reduce the presence of real-time threats within IoT networks. The results of different machine learning models which include the Logistic Regression, the Decision Tree, the Random Forest, the XGBoost, the AdaBoost, the Gradient Boosting, Bagging, K-Nearest Neighbors (KNN), and the Naive Bayes are compared based on some of the key performance indicators that are accuracy, precision, recall, F1-score, ROC-AUC, and log loss. Our findings indicate that ensemble algorithms, especially Random Forest, Decision Tree, and Bagging, can be more effective than other models in identifying a large number of detections with low false positives, and Random Forest offers an accuracy of 99.99%, precision of 99.96%, a recall rate of 99.96% and ROC-AUC of 99.99%. By contrast, the results of Naive Bayes were much worse, showing an accuracy rate of 74.28 %, a precision rate of 23.32% and an F1-score of 37.71. These findings underline that ensemble algorithms, in particular Random Forest, are also very successful in real-time intrusion detection on IoT systems. The given approach proves that ensemble learning, which possesses the capability to merge several classifiers, is an effective solution to enhancing the IoT safety of systems.
Ammar Adel Ahmed (Wed,) studied this question.