Physiological Signals Predict Exam Outcomes Using Machine Learning.

This research investigates the potential of machine learning models to forecast academic performance in exams by analyzing physiological data. During examination sessions, stress indicators such as electrodermal activity, heart rate, and skin temperature were collected and analyzed to understand their correlation with student outcomes. A diverse range of machine learning techniques was employed, from established models like logistic regression, random forest, and support vector machines, to advanced deep learning architectures including transformers, LSTMs, and GRUs. This broad approach aimed to effectively capture the intricate relationships within the physiological datasets. A particular focus was placed on evaluating the adaptability and performance of transformers when processing numerical data in this novel context. The experimental results, assessed using standard metrics like accuracy and F1-score, indicated that while deep learning models are generally adept at identifying complex patterns, simpler models like random forests sometimes achieved superior performance with added benefits of computational efficiency and interpretability. Transformers also showed comparable versatility to LSTMs and GRUs. The study emphasizes the importance of exploring a wide array of models to balance precision, efficiency, and interpretability, ultimately contributing to understanding student stressors and enhancing student well-being.