Shapley Explainable Deep Learning based Knowledge Distillation Framework for Student’s Performance Prediction

Educational institutions are constantly striving to improve the services they provide; their goal is to have the best possible teaching staff, increase the quality of teaching, and improve their students’ academic achievement. One of the initial steps taken by educational institutions was to establish Virtual Learning Environments (VLEs). This work uses a deep Explainable distillation model on a set of unique handcrafted variables collected from VLE clickstream data to forecast atrisk students and provide strategies for early intervention in such circumstances. Initially, the input data are gathered from publicly available source and pre-processing is done to enhance the quality of inputs using Missing Value Imputation. Then, in order to handle the unbalanced data, augmentation is performed by a new Improved Synthetic Minority Oversampling Technique (ISMOTE). After augmenting the data, the student’s academic performance is predicted by proposing a novel Explainable Deep Learning based Knowledge Distillation Framework (EDL_KDF). Furthermore, the proposed model can be explained using Shapley Additive Explanations (SHAP)-based reasoning. Educational institutions can evaluate, audit, and validate explanations for such relevant features more easily than explanations from DL models. The proposed framework attained 99.6 % accuracy and 99.2 1 % of precision over the Open University Learning Analytics Dataset (OULAD). Thus, the proposed study effectively predicts the students’ performance compared to other existing models.