ROBOLEARN: FACILITATING ADAPTIVE LEARNING WITHIN AN ARTIFICIAL INTELLIGENCE FRAMEWORK

Abstract

The range of student skills, preferences, and learning rates has led to a considerable increase in the need for individualized learning solutions in recent years. Using the RoboLearn environment, this work introduces an intelligent adaptive learning system that incorporates Proximal Policy Optimization-based Reinforcement Learning (PPO-RL) as its central decision-making mechanism. The goal is to provide a dynamic AI-based environment that can continually modify instructional tactics to accommodate each learner's changing demands. Experiments were carried out utilizing the Enrolled Students Dataset (2023/2024) from the Oman Open Data Portal, which has comprehensive data on learner engagement, academic achievement, and advancement, in order to assess the efficacy of the suggested PPO-RL strategy. Three popular baseline reinforcement learning models—Q-Learning, Deep Q-Network (DQN), and Asynchronous Advantage Actor-Critic (A3C)—were compared to the PPO-RL method.
Both regression-based assessment measures and classification-based metrics were used to evaluate the performance. PPO-RL produced better quantitative findings, with an F1-Score of 91.7%, Accuracy of 92.5%, Precision of 91.0%, and Recall of 92.3%. Its great predictive capacity was shown by the fact that it had the lowest Root Mean Squared Error (RMSE) at 5.88, the lowest Mean Absolute Error (MAE) at 4.12, and the highest R2 score of 0.94. Additionally, PPO-RL outperformed current models in terms of cumulative reward per episode and flexibility to unseen learners, demonstrating quicker policy convergence and needing fewer training episodes to attain optimum strategy. Visual confirmation of PPO-RL's resilience across key performance metrics was provided via a confusion matrix analysis and heatmap. Furthermore, an ablation research demonstrated the significance of distinct PPO elements including entropy bonus and policy clipping, in attaining performance stability. All things considered, this research shows that PPO-RL in RoboLearn offers a strong and expandable real-time adaptive learning system. The suggested approach is ideal for contemporary digital education platforms as it enhances learning outcomes while streamlining the delivery of teaching.This study was financially supported by the University of Applied Sciences and Technology, Ibra.