IoT Intrusion Detection Using Modified Random Forest Based on Double Feature Selection Methods

One of the fast-expanding technology today is the Internet of Things (IoT). It is very necessary, to protect these machines from adversaries and unwanted entry and alteration. Intrusion Detection Systems (IDS) are techniques that can be used in information systems to monitor identified threats or anomalies. The challenge that arises is that the IDS should detect attacks on time in high-speed network traffic data. This paper proposed a modified IDS in IoT environments based on hybrid feature selection techniques for the random forest that can be used to detect intrusions with high speed and good accuracy. IoTID20 dataset is used which has three target classes which are the binary class as normal or abnormal and the classes of categories and sub-categories for the binary class. The highest-ranked attributes in the dataset are selected and the others are reduced, to minimize execution time and improve accuracy, the number of trees in the random forest classifier is reduced to 20, 25, and 20 for binary, category, and sub-category respectively. The trained classifier is then tested and achieved accuracy approaches 100% for the binary target prediction, 98.7% for category and accuracy ranges from 78.1% to 95.2% for the sub-category target prediction. The proposed system is evaluated and compared with previous ones and showed its performance.