In modern outcome-based education practices, knowledge tracing technology has emerged as an essential tool for comprehending and optimizing the student learning process. However, existing knowledge tracing models exhibit deficiencies in terms of prediction accuracy and efficiency, as well as limited generalizability and interpretability. This paper introduces a deep knowledge tracing model based on sparse attention (DKTM-SA). Comparative experiments on several public datasets, including Algebra2005, validate that this model enhances prediction accuracy, significantly reduces the computational cost of processing long interaction sequences, and effectively aligns with learners’ cognitive development patterns.