Knowledge tracing (KT) is crucial for smart education as it analyzes students’ historical exercise-answer data to predict the accuracy of their future responses, thereby enabling teachers to gain precise insights into students’ learning statuses. However, existing models primarily focus on modeling the sequence of exercise-answer and often neglect other influencing factors in the learning process, such as knowledge point, exercise, and learning forgetting factors. Empirical evidence from educational research indicates that these factors exert a significant influence on the learning process. In this paper, we propose a Multiple learning Factors Fusion Knowledge Tracing model (MFFKT), which extracts and integrates effective features from multiple learning factors. Specifically, to accurately simulate students’ states during exercise-answer, the forgetting law is first modeled and integrated into the sequence of interactions. Then, multiple learning factors are embedded through a vectorization process. Next, static and dynamic features, as well as the temporal dependencies inherent in exercise-answer sequences, are extracted and integrated by the designed modules. Meanwhile, we simulate interference factors using adversarial samples to enhance the model’s robustness. Experimental results on real datasets demonstrate that, compared to existing KT models, MFFKT more effectively tracks students’ knowledge states and achieves superior prediction performance.