The Intelligent Tutoring Systems (ITSs) are widely used, particularly in the context of the growing prevalence of online learning. A significant challenge in ITSs is performance prediction, specifically the ability to answer correctly at the first time, commonly referred to as CFA (Correct at First Attempt). This criterion serves as one of evaluation measures for learners’ understanding and knowledge acquisition, enabling educational managers and teachers to adjust their teaching methods effectively to enhance learning outcomes. Because of the similarity between the prediction problem of learning performance and the session-based recommendation systems, this study proposes to experiment the application of session-based recommendation models, specifically LSTM (Long Short-Term Memory) in CFA prediction. In this research, two educational datasets, namely KDDCup 2010 and Assistment 2017, were employed for analysis and experimentation. The results indicate that the LSTM model outperforms other models, exhibiting a lower root mean square error (RMSE) value. Consequently, applying session data processing models to ITSs shows promise and potential for enhancing ITSs’ performance.