Comparative Analysis of Machine Learning Algorithms for IoT Security Threat Detection
Abstract
With the advent of Internet of Things (IoT) devices, which are being adopted at a very fast rate, cyberattacks will certainly be targeted and thus, intelligent intrusion detection systems that keep track of rogue activities within a network must be adequate in keeping track of such activity with high precision. Conventional signature-based methods to security are not effective in most cases to detect evolving and hitherto unknown attacks, and machine learning-based intrusion detection has a bright future. The article will compare and contrast three trained machine learning algorithms, they include Decision Tree (DT), Support Vector Machine (SVM) as well as Artificial Neural Network (ANN) in Thombing IoT threat detection of security. The experiments have been implemented with the help of the IoTID20 benchmark dataset that includes both benign and malicious network traffic that could model a variety of IoT attacks. Recursive Feature Elimination (RFE) with Random Forest was used to only select the most informative network traffic features, and then model training was done. The evaluations of the classifiers were based on Accuracy, Precision, Recall and F1-score, Areas Under the ROC Curve (AUC) and time to remove this analysis. Experimentally, it was determined that the Artificial Neural Network performed best overall with the following accuracy, precision, recall, F1-score, and AUC, respectively; 99.08%, 98.89%, 99.15%, 99.02% and 0.994, as compared to the Artificial Neural Network structures of the Decision Tree and the Support Vector Machine. The results prove that algorithmic innovation of feature selection, configured as a Random Forest, and machine learning can drastically improve an intrusion detection rate, and can be used as the solution to the protection of the modern IoT environment.