A Network Intrusion Detection Model for IoT Networks

Abstract

The integration of processing power into everyday objects has given rise to internet of things (IoT). While this is good for the actualization of ubiquitous computing, it also increases the entry points of threats into the network. This study aims to detect network intrusion attacks using three machine learning models, namely Random Forest, Decision Trees, and K-Nearest Neighbors (KNN) on two different datasets, CIC-IDS2017 and CSE-CIC- IDS2018. After pre-processing the data by removing missing values, handling outliers, and scaling features, the datasets were split into training and testing datasets in a 75:25 ratio. The classifiers were trained individually, and their performance was evaluated using various evaluation metrics such as accuracy, precision, recall, and F1-score. The best performing model was selected based on the evaluation metrics and used for making predictions on the test dataset. Random Forest achieved an accuracy rate of 0.99 on CIC-IDS2017 and 0.98 on CSE-CIC-IDS2018, while Decision Trees achieved an accuracy rate of 0.99 on CIC-IDS2017 and 0.99 on CSE-CIC-IDS2018. KNN achieved an accuracy rate of 0.99 on CIC-IDS2017 and 0.98 on CSE-CIC-IDS2018. Decision Tree algorithm was also observed to have the fastest training and testing times compared to Random Forest and KNN on both datasets. However, the Random Forest and Decision Trees models achieved similar accuracy rates, with Random Forest performing slightly better on the CIC-IDS2017 dataset and Decision Trees performing slightly better on the CSE-CIC-IDS2018 dataset. KNN had the slowest training and testing times among the three models, but still achieved high accuracy rates. These results indicate that Random Forest and Decision Trees were more accurate and faster than KNN in detecting network intrusion attacks on these datasets.