Model Evaluation of A Prediction System For Student Accreditation Using Deep Learning Algorithms Applied To Tracking Records Within The Moodle Platform.

This study presents a deep learning–based approach for predicting student academic accreditation using daily interaction logs (including weekends and holidays) from the Moodle learning management system. The dataset com-prises records from first-year students enrolled in a leveling course at the Universidad Técnica de Manabí across four academic periods between 2023 and 2025. Student behavior was modeled as temporal sequences representing the frequency of access to key Moodle components, including quizzes, fo-rums, tasks, and system interactions. Three deep learning architectures were evaluated: Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and one-dimensional Convolutional Neural Networks (CNN 1D). The mod-els were trained and evaluated using a binary classification framework, where the target variable was the accreditation status of the course. Experi-mental results demonstrate that all models achieved high predictive perfor-mance, with accuracies above 97%. Among them, the LSTM model obtained the best results, reaching an accuracy of 98.38%, highlighting the relevance of capturing long-term temporal dependencies in student interaction data. The findings confirm that patterns of engagement within Moodle constitute reliable predictors of academic success and support the potential of deep learning techniques for early identification of students at risk, enabling time-ly pedagogical interventions in higher education contexts.