RF-SVM: Robust Injection Attack Detection Using Random Forest Feature Selector and Support Vector Machine Classifier for Secure Network Systems

Abstract

IoT is rapidly growing and being implemented in various domains, such as smart cities, smart governments, smart hospitals, and smart homes. However, this growth also brings security challenges that can have detrimental effects on the IoT system. Injection attacks are one example of these challenges, which can compromise the network's performance and functionality. This paper focuses on injection attacks and uses machine learning methods to detect these attacks and suspicious activities in the system. The main challenge is the availability of much missing data in the corresponding datasets. Contrary to previous research, which is based on eliminating the sparse features, this paper proposes a method for keeping such sparse, valuable data. The proposed method combines a support vector machine with features selected by the random forest to detect this type of attack with high confidence. The performance was evaluated using metrics such as accuracy, precision, and recall. The proposed model offers low computational overhead and high processing speed, making it well-suited for various network devices, including smart homes. Finally, by selecting 46 effective features from the total feature set of the AWID dataset, we achieved a detection accuracy of 99.99%. To the best of our knowledge, this study has had the best performance compared to previous studies.