An efficient ant colony-driven hierarchical attention sparse autoencoder for intelligent digital education quality evaluation

The rapid expansion of digital education has created an urgent need for intelligent evaluation systems capable of assessing learning quality accurately and efficiently. Existing approaches often rely on conventional evaluation frameworks, which struggle with high-dimensional, heterogeneous, and unbalanced datasets, leading to limited assessment accuracy. This study proposes a Deep Learning (DL)-based framework, the Modified Ant Colony-driven Hierarchical Attention Sparse Autoencoder Network (MAC-HASA), to address these challenges. Hierarchical Attention focuses on the most relevant features in high-dimensional digital learning data. Sparse Autoencoder helps to reduce feature redundancy while learning compact representations of the data. MAC Optimization used to dynamically tune feature weights to highlight informative signals and improve model convergence The framework integrates sequential data preprocessing, including z-score normalization and dimensionality reduction through Robust Principal Component Analysis (RPCA), with mutual information-guided embedding for effective feature extraction across heterogeneous data. Experimental evaluation on a Digital Education Quality Evaluation Dataset containing 2000 learner records demonstrates that MAC-HASA outperforms conventional DL baselines, achieving an accuracy of 0.959, precision of 0.945, recall of 0.937, and F1 score of 0.941. These results indicate the model’s capability to capture complex learning interactions, provide timely and reliable predictions, and enhance interpretability. Overall, MAC-HASA offers a robust and scalable solution for intelligent digital education assessment, providing actionable insights to educators and administrators for improving learning outcomes and optimizing online learning environments.