Automated Grading Systems and Assessment for Open Book Examinations in Higher Education using Graph Neural Network Models

The COVID-19 pandemic has caused several universities to switch to online or hybrid teaching approaches. In contrast to traditional in-person exams, which provide controlled oversight, online assessments present issues for upholding academic integrity since they allow students to interact and exchange materials without supervision. Research on possible grading schemes to lessen these difficulties has been sparked by open-book exams in higher education. Data preparation, feature extraction, and model training are the three phases of the suggested system. Data goes through a number of cleaning steps during preprocessing, and algorithms are used to detect and evaluate developing tendencies during feature extraction. GNNs were used because of their relational connection handling capabilities, however their dependence on preset graph structures is problematic when dealing with multivariate data. With an impressive accuracy rate of 95.28%, the system demonstrated its effectiveness in evaluating student performance. These findings show how DL models can improve the way open-book exams are evaluated in grading schemes. By using feature extraction techniques and thorough data pretreatment processes, institutions increase the validity and reliability of online exams. The method lessens the effects of academic dishonesty and lays the groundwork for maintaining integrity in modern classrooms.