Knowledge tracing (KT), which involves the use of students' problem-solving histories to estimate their current level of knowledge, has attracted significant interest from researchers. Modeling learners' knowledge state can be used to recommend appropriate learning content; consequently, it can serve as a foundational technology to support personalized learning. Nevertheless, using only student problem-solving data is subject to certain limitations when attempting to model their knowledge states. Recent studies have attempted to address this issue by incorporating various side information in terms of questions, knowledge components, and students. However, the majority of the side information used in prior studies was constrained to the process of problem solving, thereby lacking the ability to adequately capture information from the students' side. In the real world, students engage in various learning activities that significantly impact the changes in their knowledge states. Therefore, students' learning behaviors (SLBs) during the online learning process can be useful in estimating their knowledge states. In this study, we propose SLB-KT that uses SLB data. We collected learning behavior and problem-solving data from the same students from an e-learning platform. Following a literature review, learning styles and interaction types were adopted as the background theories for this study. Student behavior modeling was then performed in two dimensions: contextual information based on learning style and interaction type, and behavior pattern. Subsequently, we demonstrated the performance of the SLB-KT models in many aspects using various baseline models. The experiment showed that the SLB-KT models outperformed the baseline models in terms of prediction performance, alleviated the cold-start issue, and improved reliability.