Perbandingan Kinerja Model Transformer dan LSTM dalam Prediksi Tingkat Pemahaman Siswa terhadap Materi Pembelajaran Digital

Abstract

The rapid adoption of digital learning platforms has generated large volumes of student interaction data that can be utilized to predict students' comprehension levels and support timely educational interventions. This study aims to compare the performance of the Transformer and Long Short-Term Memory (LSTM) models in predicting students' comprehension of digital learning materials. A quantitative comparative experimental approach was employed using student activity data collected from a Learning Management System (LMS). Data preprocessing included cleaning, normalization, sequence generation, and feature selection using Particle Swarm Optimization (PSO) to identify the most relevant features before model training. Model performance was evaluated using Accuracy, Precision, Recall, and F1-Score on the same testing dataset. The results indicate that the Transformer model outperformed LSTM across all evaluation metrics, achieving an accuracy of 92.3% and an F1-Score of 92.0%, while LSTM achieved an accuracy of 88.5% and an F1-Score of 87.9%. These findings demonstrate that the self-attention mechanism enables the Transformer model to capture complex relationships among learning features more effectively than LSTM. Therefore, the Transformer model is recommended for developing adaptive learning systems capable of accurately predicting student comprehension and supporting early academic intervention.