APP sensitive behavior surveillance method for smart terminals based on time series classification

Abstract

Security mechanisms based on permission are mainly applied to smart terminals in industrial Internet of Things (IoT), which restrict the underlying behavior of tracking, Application Programming Interfaces (APIs) analysis and resource scheduling on non-root devices. In order to achieve non-invasive sensitive behavior surveillance, this article studies the mapping between time series data and sensitive behavior of applications (APPs). By triggering APP behavior, multivariate time series data of system state can be automatically collected through side channel scanning process, then data labeling, data slicing, and model training are carried out. After that, a sensitive behavior surveillance method based on time series classification through deep learning is explored. Experimental results show that compared to other baseline models, the proposed behavior surveillance based on convolutional neural networks (CNNs) exhibits performance advantages and robustness, where the overall accuracy of the test set in mixed behavior scenarios exceeds 70%.