Bayesian Model for Academic Performance Prediction in Learning Analytics

Abstract

The learning analytics dashboard (LAD) enables the prediction, tracking, and early recommendation of actions based on academic performance, student conduct, cognitive abilities, and personality traits. The creation of a questionnaire, which involved expert validation and a pilot test, came before the data gathering. Students independent traits, such as behaviour, cognitive skill, and personality, and academic data are the contingent variables. The study flow comprises knowledge acquisition, data collection and analysis, implementation that executes experiments, and evaluation. An exploratory data analysis (EDA) has been performed to investigate the patterns and trends within the data collected on students' characteristics. Simultaneously, a Bayesian prediction model has been created and trained using the gathered data. The forecasts yielded an accuracy of 81%. Additionally, the F1 scores of 0.74 indicate a moderate level of ability, while scores of 0.90 suggest a high level of performance, and scores of 0.98 indicate an excellent level of performance. The interpretation of the F1 score is conducted within the specific realm of the issue and compared to manual calculations. A satisfactory level of performance is considered acceptable. Within the scope of this investigation, a sensitivity score of 0.8 is deemed favourable due to its ability to accurately identify the related risk of misclassification.