Publication Details
Abstract
Schools face two key challenges: predicting the academic outcomes of their students and preventing identity impersonation in online educational systems. In this article, an integrated system, which includes explainable artificial intelligence for predicting students' achievements and iris-based biometric authentication, is presented to provide a secure learning environment. Our solution uses a Random Forest classifier trained on 1,044 records of students based on 33 demographic, academic, and behavioural characteristics to predict student performance at an accuracy of 89.5%. The feature of AI explainability uses SHAP values and permutation importance analysis to identify major predictors and identifies the following as the most significant factors in determining student outcome: past academic performance (G2, G1) and history of failure. For biometric authentication, we employ a Vision Transformer (ViT) model that is trained on synthetic iris data with a zero false acceptance rate against unknown individuals and a high recognition rate against enrolled students. The fusion of these systems offers a comprehensive framework where academic predictions are securely linked to authenticated student identity, preventing impersonation and ensuring data integrity. Experimental testing demonstrates the efficacy of the system in real-world scenarios with explainable outcomes but secure processes, greatly cherished by teachers. This is the first full incorporation of explainable student analysis and biometric anti-fraud strategies, which holds vast potential for use in current-day schools requiring analytical outcomes and assurance of security.