Evaluating the Effectiveness of Deep Learning Models in Student Performance Prediction

Abstract

New sources of educational data are becoming more accessible, permitting the use of Artificial Intelligence to track
and forecast students' performances, therefore allowing prompt actions to be taken to improve their learning
experiences. This study analyzes the application performance of deep learning architectures, including
Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNN), and Long Short-Term Memory
(LSTM) networks, in predicting students’ academic performance. The models were trained and evaluated on a
dataset containing the demographics, attendance records, assignment scores, and exam results of 5000 students.
Each model underwent evaluation for accuracy, precision, recall, F1 score, and the overall prediction score. The
LSTM model was the most accurate among the models tested, achieving 89% accuracy, confirming its advantage in
capturing temporal dependencies within the student data. Compared to conventional machine learning approaches
like Decision Trees and Support Vector Machines (SVM), the LSTM model demonstrated markedly enhanced
performance. The results demonstrate the power latent in the use of deep learning for educational data mining,
advocating for its greater use in intelligent tutoring systems and broader learning analytics frameworks. This work
adds to the existing literature by providing evidence and methodological frameworks regarding model choice and
deployment strategies aimed toward optimizing educational outcomes through automated systems.