User Modeling via Anomaly Detection Techniques for User Authentication

Anomaly detection is quickly becoming a very significant tool for a variety of applications such as intrusion detection, fraud detection, fault detection, system health monitoring, and event detection in IoT devices. An application that lacks a strong implementation for anomaly detection is user trait modeling. User trait models expose up-to-date representation of the user so that changes in their interests, their learning progress or interactions with the system are noticed and interpreted. The reason behind the lack of adoption in user trait modeling arises from the need for a continuous flow of high-volume data, that is not available in most cases, to achieve high-accuracy detection. This paper provides new insight into the anomaly detection techniques through Big Data utilization. With Big Data characteristics, i.e., volume, variety and velocity, anomaly detection techniques have become more suitable tools for user trait modeling. User traits will be modeled by creating a security user profile for each user. This profile is structured and developed to be a source for a strong real-time user authentication method. An ingenious implementation of three models; k-means, HMM, and auto-encoder neural network has been presented that automatically and accurately build a unique pattern of the users' behavior. The implementation comprises four main steps: prediction of rare user actions, filter security potential actions, build/update a user profile, and generate a real-time (i.e. just in time) set of challenging questions. Real-world scenarios have been given showing the benefits of these challenging questions in building secure knowledge-based user authentication systems.