Knowledge Tracing (KT) is a technique that predicts students’ future performance based on their historical exercise records. A growing body of research has recognized the value of student behavioral features in KT modeling. However, existing approaches often struggle to flexibly incorporate these features due to model complexity, limiting their ability to adaptively select and integrate the most informative behavioral signals. To address these limitations, this study proposes a Behavior-Enhanced Dynamic Graph Knowledge Tracing (BEDGKT) model. This work employs a two-layer MLP to capture student behavioral features, significantly reducing feature engineering efforts. Furthermore, we introduce the concept of student behavioral preference and leverage a double-layer GRU architecture to fully utilize behavioral feature information. The experiments were conducted on three public datasets, demonstrating AUC improvements of 1.2% – 5.4% across different datasets. Through visualization techniques and case studies, we quantitatively validated the impact of behavioral features on prediction performance. This work provides a more scalable and practical solution for knowledge tracing, offering both theoretical significance and practical implications for advancing personalized education.