Enhancing MOOC Course Classification with Convolutional Neural Networks Via Lion Algorithm-Based Hyperparameter Tuning

The trend toward distance education in higher education, which was accelerated in adoption by COVID-19, has amplified the requirement for personalized course recommendations and accurate classification of Massive Open Online Courses (MOOCs). However, many frameworks use static, rule-based systems that are hard to change and do not allow for diverse course structures. This paper proposes a hybrid technique for the classification of MOOCs courses based on the integration of Convolutional Neural Networks (CNN) and the Lion Optimization Algorithm (LOA). It ensures better performance in classification by including the strongest pattern recognition capability of CNN and the powerful hyperparameter optimization by LOA. We also contrasted our approach with various algorithms, including biologically motivated optimization algorithms like the Whale Optimization Algorithm (WOA), Particle Swarm Optimization (PSO), and Ant Colony Optimization (ACO). We also contrast deep learning architectures like Bidirectional Encoder Representations from Transformers (BERT) and Long Short-Term Memory (LSTM) and transformer-based embeddings (sentence embeddings from a pretrained transformer (SimCSE) plus an MLP classifier). We conduct our evaluation on a new dataset that has not been studied in the literature. These experimental results show that the CNN/LOA outperforms the other methods, with testing accuracy at 96.34