Tausia Tahsin Nuzum, Sumita Tasnim, Tasnia Rafique Simran, Nabarun Halder, Ashraful Islam, M Ashraful Amin
2025 IEEE 23rd Student Conference on Research and Development (SCOReD)
In: 2025 IEEE 23rd Student Conference on Research and Development (SCOReD)
IEEE, pp. 1-6
Social media use is prevalent among students and is associated with procrastination that can negatively affect academic performance. Detecting at-risk students from small, imbalanced surveys is challenging. This study examines whether synthetic data can improve prediction. Risk prediction was performed using seven classifiers: logistic regression (LR), random forest (RF), gradient boosting (GB), support vector machine (SVM) with RBF kernel, k-nearest neighbors (KNN), Gaussian Naive Bayes (GNB), and a multi-layer perceptron (MLP), on two small student surveys. Data augmentation techniques used are: Synthetic Minority Over-sampling Technique for Nominal and Continuous Features (SMOTENC), Gaussian Copula (GC), and Conditional Tabular Generative Adversarial Network (CTGAN). Without augmentation, SVM, GB, and RF achieved similar accuracy on Dataset-1 (≈0.93). GNB led Dataset-2 (0.907 accuracy), with RF, GB, SVM, and MLP following closely (≈0.89-0.90). Augmentation effects are model and dataset dependent and generally modest with Tree ensembles remaining largely stable. The model with the largest improvement was RF, gaining a +0.013 increase in F1 on GC for Dataset-2. The MLP declined with synthetic data. Hence synthetic augmentation might not be a universal remedy for small-scale datasets. Overall, lightweight models trained on a modest set of smartphone-like features can predict the risk of procrastination affecting academic performance and are suitable for on device use.
Assistant Professor
Department of Computer Science and Engineering
Independent University, Bangladesh (IUB)Human-Computer Interaction, AI for Social Good, AI for Public Health, AI for Impact