2025 ASEE Annual Conference & Exposition

Large amount of data is obtained from online courses, e-learning platforms, and institutional technologies. Educational Data Mining (EDM) leverages these data to make informed decisions that improve the educational experience, student outcomes, and institutional efficiency. As academic achievement is one of the key aspects for assessing quality education, predicting students' academic performance has become a focus for research in this field. Several traditional machine learning models like Naïve Bayes, K-Nearest Neighbors (KNN), Logistic Regression and a few more are most widely used in EDM field to predict students' academic performance. Though Convolutional neural networks (CNNs) are widely used in several other domains, they have not been extensively studied for educational data mining. In this study, we proposed a novel hybrid model that combines the CNNs' feature extraction strength with the traditional classification model. We first converted our 1D numerical student data into 2D image representations. This allows 2D-CNN applications to extract important features from the data. The extracted features are then fed into traditional classification models, including Naïve Bayes, K-Nearest Neighbors, and Logistic Regression. The performance of the hybrid model is evaluated in a pass/fail classification scenario. Experimental results show that our proposed CNN-based hybrid classification model outperforms the standalone traditional model in terms of classification accuracy. This study introduces an innovative approach in the educational domain, demonstrating that CNNs can provide a more robust and reliable method for predicting student performance, especially when predicting binary results like pass or fail.