Abstract: Knowledge tracing plays a crucial role in the process of intelligent education. It involves modeling the evolution of learner's knowledge state over time based on their historical exercise sequences and predicting their future performance. Although existing knowledge tracing methods, particularly those based on Transformer architecture, have achieved significant predictive performance, most of these methods rely solely on the learner's response results for modeling. They ignore other important feature information generated during the exercise process, especially features related to the learner's ability, and fail to consider the loss of original features in the process of model prediction. To solve the above issues, a novel Feature-Enhanced Attentive Knowledge Tracing model (FE-AKT) is proposed. Firstly, the model introduces learner-specific features such as learner's response time to construct ability features, providing richer feature representations for attentive knowledge tracing. Then, learners with similar abilities are clustered into the same group to be incorporated into the model training. Additionally, during the model training process, exercise features are repeatedly input to the model, and exercise embeddings are fused with question-answer embeddings, which enhances the representation of learner features and reduces data loss. The comparative experimental results on real online education datasets demonstrate that compared with classical models and recent state-of-the-art models, the proposed model exhibits strong performance predictive capabilities, and it demonstrates good interpretability.