Real-Time Behavioral Biometric Information Security System for Assessment Fraud Detection

Abstract

Online education has become a major delivery method in education. Many schools have adopted online delivery of courses. This has exposed schools to greater information security risks, such as assessment cheating, identity theft, and loss of sensitive student information. To counter emerging new types of attacks, risks, and vulnerabilities, the protection measures must be able to adapt to the evolving behaviors of both user and attacker by learning new emerging vulnerabilities and behaviors of users. We propose a new Real-time Behavioral Biometric Information Security (RBBIS) architecture that non-invasively builds behavioral profiles on the fly using deep-learning approaches. The method learns behaviors of students to validate users and detect intrusion, identity theft, and assessment fraud. This greatly improves the current limitations of the existing user authentication approaches of online education platforms. RBBIS was evaluated using CNN deep-learning keystroke behavior biometric analysis and compared with various existing machine learning algorithms: j48, Naive Bayes, and Multi Layered Perceptron (MLP). The results show that our deep-learning method performed best with a Convolutional Neural Network (CNN) with 92.45% of accuracy, whereas Naive Bayes, j48 and MLP achieved accuracies of 68.87%, 73.38% and 77.11%, respectively.