From stress to scores: a machine learning analysis of wearable sensor data during examinations for predicting academic performance from physiological stress signals

Many students suffer from heavy stress when it comes to their exams, often disabling them from performing their best due to a fear of failure. This is significant because it affects one’s health, grades, and future endeavors. But how does exam stress affect students’ exam performance? Our objective is to decipher which physiological indicators affect the exam score the most. This study investigates whether physiological signals captured by wearables—electrodermal activity (EDA), heart rate (HR), blood volume pulse (BVP), skin surface temperature (ST), inter-beat interval (IBI), and accelerometer (ACC) data—can predict exam scores for three testing events (Midterm 1, Midterm 2, Final). To assess this, we used signal processing and feature importance in measuring and predicting the exam performance. I found that the feature that affected exam scores the most was the BVP readings, also known as the Blood volume pulse amplitude, had an importance of 0.068433.