To use an intelligent tutoring system (ITS) in an educational setting effectively, it is necessary to understand the skill status of students and recommend appropriate questions. Existing studies focused on improving the performance of ITS using student modeling to estimate the skill status of students. Knowledge tracing (KT) is the mainstream student modeling method, and deep learning approaches such as deep knowledge tracing and self-attentive knowledge tracing have been studied extensively in recent years. These models take only the questions solved by the student and the correct or incorrect answers to those questions as input; they do not assume the use of other features. In this study, we perform student modeling using DeepFM and FiBiNET that combines factorization machines and deep learning. Our results indicate that these models are more accurate than KT because of their ability to cope with sparse data and consider pairwise feature interactions, and more suited for real-world applications than KT.