Online learning has become pivotal in modern education, providing flexible learning methods and abundant resources. However, effectively assessing students’ knowledge mastery remains challenging. Knowledge Tracing (KT) addresses this by analyzing student behaviors and tracing their knowledge states. It supports applications like course recommendation and personalized teaching. Existing KT models often overlook features such as resource difficulty and student ability, which leads to significant limitations. Therefore, we propose Multidimensional Features-Based Knowledge Tracing (MFKT). This method integrates resource difficulty and student ability based on the CL4KT framework, which utilizes contrastive learning. We replace CL4KT’s transformer encoders with GRU for improved time-series feature capture and computational efficiency. We conducted experiments on three datasets using anonymized public data to protect privacy. Our results show that MFKT performs better than existing KT models. Ablation studies confirm the effectiveness of incorporating resource difficulty and individual ability.