In recent years, with the widespread adoption of smart education, Intelligent Tutoring Systems (ITSs) have played a crucial role in enhancing learning quality. Among them, Knowledge Tracing (KT) is a fundamental research area in ITS, aiming to establish the relationship between students and knowledge concepts by extracting historical response data and predicting their future problem-solving performance. However, existing KT models face limitations when handling long learning sequences in complex educational settings, often neglecting critical characteristics of student groups. To address these challenges, this paper proposes the improving Knowledge Tracing leveraging student group classification with Optimized-Transformernet model (OTKT). OTKT incorporates a student group classifier, effectively accounting for individual differences among students and enhancing interpretability. Additionally, it employs interaction sequence sampling to explore question discriminability, providing the model with more targeted input data. By refining the Transformer architecture, the model improves the ability to capture long-range dependencies in knowledge states during attention computations while maintaining a balanced trade-off between model complexity and computational efficiency. Experiments on four datasets against baseline models demonstrate the efficiency and robustness of OTKT.