Adaptive artificial intelligence framework for skill level prediction and music training recommendation

Abstract

Accurate assessment of learner skill levels and timely adaptation of training programs remain critical challenges in music education. The increasing availability of learner performance data creates opportunities for intelligent systems to support personalized training and continuous skill development. Conventional music training systems often depend on rigid curricula and static evaluation methods, which fail to account for nonlinear performance patterns, learning dynamics, and individual differences, leading to insufficient adaptability and less effective training recommendations. The objective is to design an intelligent system capable of predicting student performance, an Adaptive Performance-Driven Training System (APDTS) is introduced, integrating deep learning–based performance prediction with adaptive reinforcement learning–based decision making. The framework operates on structured learner profile data, performance metrics, and session-wise practice records collected over multiple learning cycles. Data quality and numerical stability are ensured through missing value imputation and Z-score standardization. A convolutional neural network performs joint feature learning and performance prediction. The predicted performance indicators are subsequently utilized as state inputs for an Adaptive Double Deep Q-Network enhanced with a Resilient Enzyme Action Algorithm (A2DQN-REA), enabling stable and adaptive training program recommendation. The experimental results showed that APDTS outperformed standard approaches in terms of ROC-AUC (0.953), Accuracy (0.975), Precision (0.970), Recall (0.972), F1-Score (0.984) and lower MSE (0.029). The APDTS effectively combines performance prediction and adaptive decision-making to deliver personalized training strategies, offering a robust and scalable solution for intelligent music learning environments.