Least Squares-Based Multidimensional Student Achievement Prediction for Smart Education

Abstract

Traditional teaching methods can no longer meet current management needs. The use of modern scientific and technological means to assist decision-making and modernization is imperative. Student performance is an important basis for evaluating the quality of teaching and learning, as well as a key indicator of students’ mastery of knowledge. This paper proposes a data-driven method for the predictive analysis of academic performance, introducing a novel Orthogonal Kernel Least Squares (OKLS) model to predict students’ exam outcomes. The core innovation of the OKLS method lies in its integration of kernel-based nonlinear mapping with an orthogonal forward selection sparsification strategy. This approach systematically constructs a sparse and interpretable model by selecting the most significant basis vectors, thereby effectively mitigating overfitting and reducing computational complexity. Compared to Support Vector Machine (SVM) and Random Forest (RF) models, OKLS demonstrates comprehensive advantages. Experimental results on a real-world student dataset show that OKLS achieves superior predictive accuracy in both classification (pass/fail) and regression (specific score) tasks. Furthermore, the OKLS model exhibits significantly higher sparsity than the compared methods, resulting in a more computationally efficient and interpretable solution without compromising predictive power. The findings validate OKLS as a robust and efficient framework for student performance prediction.