Intrusion Detection System Using Hybrid Machine Learning Classifiers and Optimum Feature Selection in Internet of Things (IoT)

Abstract

This paper has harnessed the extensive IOT2023 dataset, encompassing both legitimate and malicious IoT data, for the purpose of training and testing machine learning classifiers. The performance of the Decision Tree Classifier has been even astonishing: 99% precision, 99% recall, and 99% accuracy. Summarily, the Decision Tree and the Random Forest classifiers have performed exceptionally, where the Random Forest had an upper hand, especially in identifying benign cases precisely, with 99% accuracy and precision. The K-Nearest Neighbors (KNN) classifier has also performed as good, with high precision and recall that gave a 99% accuracy and 92% precision. The choice of the best classifier will be able to depend on the unique needs of an application and in consideration of different valid performance metrics. There is a possibility of further refinement and fine-tuning that may bring these classifiers to a higher level of performance. Therefore, this work will further the state of the art in intrusion detection but also underline the very importance that so far does not receive enough attention in the real-world application: careful selection and refinement of models. Among them, the Random Forest has reached a 99% accuracy, 99% precision, 93% recall, and 84% F1-score. KNN has reached 99% in accuracy, 92% in precision, 82% in recall, and 84% in F1, which has proved to be a robust algorithm to provide a safe IoT ecosystem.
Copyright © 2024 Praise Worthy Prize - All rights reserved.