Uncertainty-Aware Bayesian Adaptation Improves Reliability of Student Psychological Assessment Tools

Abstract

The assessment of students' psychological status is one of the major aspects for detecting problems in terms of students' mental health at school which influence the performance of learners. However, there have been improvements in application of AI methods to achieve accurate assessment. However, the current assessment models fail to provide an assessment of prediction uncertainty which makes them less effective in the educational context. In order to solve this problem, an Uncertainty-Aware Bayesian Adaptation (UABA) framework is suggested in the paper, and it includes data preprocessing, posterior estimation of the Bayes model, adaptive learning, and confidence-aware classification in order to deliver a reliable student psychological assessment. The novel thing about the UABA framework is that in addition to predicting the psychological condition of students, it provides a measure of confidence of each prediction with the help of Bayesian uncertainty estimation. The Bayesian adaptive mechanism helps optimize the parameters of the model continuously based on the incoming data, thereby reducing overconfident incorrect predictions and decision-making. The UABA framework has been applied to a benchmark student psychological assessment dataset and benchmarked with different machine learning and deep learning models such as SVM, Random Forest, CNN, LSTM and Transformer. Experimental results revealed that the proposed UABA model had a maximum accuracy of 97.82%, precision of 97.56%, recall of 97.31%, F1 score of 97.43% and AUC of 98.11% among all the comparative models. In addition to this, the false positive rate, false negative rate, prediction confidence, and calibration score of the framework were 2.18%, 2.69%, 97.92%, and 98.26%, respectively, achieving better than the other models in terms of prediction performance and uncertainty estimation. These results further support the viability of uncertainty aware Bayesian adaptation as a reliable methodology in psychological evaluation of students with the use of AI.