Machine Learning Models for Secure Data Analytics: A taxonomy and threat model

In recent years, rapid technological advancements in smart devices and their usage in a wide range of applications exponentially increases the data generated from these devices. So, the traditional data analytics techniques may not be able to handle this extreme volume of data known as Big Data (BD) generated by different devices. However, this exponential increase of data opens the doors for the different type of attackers to launch various attacks by exploiting various vulnerabilities (SQL injection, OS fingerprinting, malicious code execution, etc.) during data analytics. Motivated from the aforementioned discussion, in this paper, we explored Machine Learning (ML) and Deep Learning (DL)-based models and techniques which are capable off to identify and mitigate both the known as well as unknown attacks. ML and DL-based techniques have the capabilities to learn from the traffic pattern using training and testing datasets in the extensive network domains to make intelligent decisions concerning attack identification and mitigation. We also proposed a DL and ML-based Secure Data Analytics (SDA) architecture to classify normal or attack input data. A detailed taxonomy of SDA is abstracted into a threat model. This threat model addresses various research challenges in SDA using multiple parameters such as-efficiency, latency, accuracy, reliability, and attacks launched by the attackers. Finally, a comparison of existing SDA proposals with respect to various parameters is presented, which allows the end users to select one of the SDA proposals in comparison to its merits over the others.