Integrating Reinforcement Learning with Dynamic Knowledge Tracing for personalized learning path optimization.

Area of Science:

• Educational Technology
• Artificial Intelligence in Education
• Machine Learning for Learning Analytics

Background:

• Personalized learning systems require dynamic adaptation to individual student needs.
• Traditional Knowledge Tracing (KT) models struggle with evolving student knowledge and task sequencing.
• Existing methods lack flexibility in adapting learning paths for optimal student engagement and retention.

Purpose of the Study:

• To propose and evaluate RL-DKT, a novel framework integrating Dynamic Knowledge Tracing (DKT) with Reinforcement Learning (RL).
• To enhance personalized learning by dynamically optimizing task selection and difficulty based on real-time student performance.
• To improve student engagement, retention, and overall learning outcomes in adaptive educational systems.

Main Methods:

• Developed the RL-DKT framework, combining DKT for knowledge state tracking and RL for dynamic task selection.
• Utilized three real-world educational datasets (ASSISTments, KDD Cup 2010, Cognitive Tutor) for evaluation.
• Compared RL-DKT against traditional KT models like Bayesian Knowledge Tracing (BKT) and DKT.

Main Results:

• RL-DKT demonstrated superior performance across multiple metrics compared to conventional KT models.
• Achieved a 12.5% improvement in task completion time and a 7.6% enhancement in prediction accuracy.
• Significantly reduced dropout rates by 50%, indicating improved student engagement and learning path optimization.

Conclusions:

• The RL-DKT framework effectively addresses limitations of traditional KT models in personalized learning.
• Dynamic adaptation of tasks and difficulty through RL significantly enhances learning efficiency and student engagement.
• RL-DKT offers a promising approach for developing more effective and adaptive educational technologies.