IoT-Malware Detection and Classification by Deep Learning Base Feature Mapping and Ensemble Learning

Abstract

With the rapid proliferation of Internet of Things (IoT) devices, which offer a myriad of ways to breach security, IoT devices have become one of the main targets for malware today. Conventional forms of detection that are based on knowledge of signatures often do not detect newly developed or obfuscated malware. In this work, we propose an ensemble learning approach for a deep learning-based malware detection and classification scheme tailored to the IoT environment. The method involves pre-processing raw network traffic data to preprocess raw network traffic data and convert it into a structured form. After that a one-dimensional Convolutional Neural Network (1D CNN) is leveraged to extract deep/middle level features that would capture the temporal and behavioral features of network traces. Finally, an ensemble of classifiers, Random Forest, Gradient Boosting, and XGBoost, among others, is applied to the computed features for the actual classification. We evaluate our method on the public-dated IoT malware dataset in our project, experiment results demonstrate that our proposed approach achieves better accuracy, precision and recall than baseline methods. The proposed architecture is highly reliable and adaptive to guarantee the effectiveness of real-time IoT protection systems with deep feature mining combined with ensemble learning which has enabled us to maintain the good performance.