Biblio

The Internet of things (IoT) is proving to be a boon in granting internet access to regularly used objects and devices. Sensors, programs, and other innovations interact and trade information with different gadgets and frameworks over the web. Even in modern times, IoT gadgets experience the ill effects of primary security threats, which expose them to many dangers and malware, one among them being IoT botnets. Botnets carry out attacks by serving as a vector and this has become one of the significant dangers on the Internet. These vectors act against associations and carry out cybercrimes. They are used to produce spam, DDOS attacks, click frauds, and steal confidential data. IoT gadgets bring various challenges unlike the common malware on PCs and Android devices as IoT gadgets have heterogeneous processor architecture. Numerous researches use static or dynamic analysis for detection and classification of botnets on IoT gadgets. Most researchers haven't addressed the multi-architecture issue and they use a lot of computing resources for analyzing. Therefore, this approach attempts to classify botnets in IoT by using PSI-Graphs which effectively addresses the problem of encryption in IoT botnet detection, tackles the multi-architecture problem, and reduces computation time. It proposes another methodology for describing and recognizing botnets utilizing graph-based Machine Learning techniques and Exploratory Data Analysis to analyze the data and identify how separable the data is to recognize bots at an earlier stage so that IoT devices can be prevented from being attacked.

With the rapid development of the mobile Internet, Android has been the most popular mobile operating system. Due to the open nature of Android, c countless malicious applications are hidden in a large number of benign applications, which pose great threats to users. Most previous malware detection approaches mainly rely on features such as permissions, API calls, and opcode sequences. However, these approaches fail to capture structural semantics of applications. In this paper, we propose AMDroid that leverages function call graphs (FCGs) representing the behaviors of applications and applies graph kernels to automatically learn the structural semantics of applications from FCGs. We evaluate AMDroid on the Genome Project, and the experimental results show that AMDroid is effective to detect Android malware with 97.49% detection accuracy.