Document Type : Original Research Paper
Authors
1 Department of computer engineering, Nat.C., Islamic Azad University, Natanz, Iran
2 Department of Computer engineering, faculty of electrical and computer engineering, technical and vocational university, (TVU), tehran, iran
Abstract
Background and Objective:
Educational data mining, as a modern and interdisciplinary field, plays a crucial role in analyzing learners’ behavior, identifying factors influencing academic success or failure, and supporting educational decision-making. With the growth of e-learning systems and the increasing volume of educational data, the need for intelligent methods to extract hidden knowledge from these data has become more pronounced. Although traditional machine learning methods have been widely used in numerous studies, they face limitations in terms of accuracy and generalization capability when dealing with complex, nonlinear, and high-dimensional educational data. In this context, deep learning–based models and artificial neural networks have attracted significant attention due to their strong ability to model complex relationships. The main objective of this study is to develop and evaluate an intelligent hybrid model based on feedforward neural networks, including Artificial Neural Networks ( ANN ), Extreme Learning Machines (ELM), and Multilayer Perceptron (MLP), combined with advanced feature selection methods, in order to identify key variables affecting students’ academic performance and significantly improve the accuracy of academic status prediction.
Research Methodology:
In this study, a comprehensive set of educational data was utilized, including demographic, socio-economic, enrollment-related information, educational records, and students’ academic grades. The data were collected from university educational systems as well as reputable public datasets. After performing necessary preprocessing steps such as data cleaning, handling missing values, and feature normalization, the feature selection process was carried out using three methods: MRMR, Chi-square, and ReliefF. Subsequently, the three neural networks (ANN, ELM, and MLP) were trained using 90% of the data, while the remaining 10% was used for testing and model validation. Finally, the outputs of the networks were integrated into a hybrid model using a majority voting strategy. The performance of the proposed model was evaluated using standard metrics, including accuracy, precision, recall, and the F-measure.
Findings and Contributions:
The experimental results demonstrated that the ReliefF method outperformed the other feature selection techniques in identifying influential features. Using the top 20 features selected by this method, the proposed hybrid model achieved an accuracy of 81.44% and an F-measure of 72.09%. In the evaluation of individual neural networks, the ELM model exhibited the best performance with an accuracy of 82.8%, which was on average 2–4% higher than that of ANN and MLP. Moreover, comparison with traditional machine learning approaches revealed that the hybrid neural network model improved prediction accuracy by more than 7% and classification accuracy by over 4%, indicating the significant superiority of the proposed approach.
Conclusion and Future Work:
The findings of this study indicate that combining feedforward neural networks with appropriate feature selection methods provides an efficient and reliable approach for predicting students’ academic status. The proposed model can serve as an intelligent decision-support tool in educational systems to enable early identification of students at risk of academic failure. Future research may focus on incorporating attention mechanisms, more advanced feature selection techniques, deeper learning models, and larger and more diverse datasets to further enhance the accuracy and generalizability of the model.
Keywords
', './data/jte/coversheet/cover_en.jpg'))
Send comment about this article