EXEDA: An Electrodermal Activity Dataset of Engineering Students During Mock Examinations for Grade Prediction

EXEDA is a publicly available dataset of continuous Electrodermal Activity (EDA) recordings collected from 65 fourth-year engineering students (55 male, 10 female; aged 21–22) during an individual, high-stakes machine learning mock examination. It was designed to support research on physiological responses to academic stress, an area where public datasets combining continuous EDA with real examination contexts and performance outcomes remain scarce. High-stakes assessments activate the sympathetic nervous system, producing autonomic responses, such as EDA, that may relate to attention, working memory, and problem-solving under stress. EXEDA offers a controlled, ecologically realistic setting to study these dynamics, combining physiological signals with self-reported psychological state and performance outcomes. Protocol. Each recording covers three consecutive phases, recorded continuously in a single file with no phase boundaries marked in the data: Baseline: 2 minutes + Examination: 15 minutes + Recovery: 2 minutes Users who need phase-level analysis should segment each file using this fixed timing (0–2 min baseline, 2–17 min examination, 17–19 min recovery), based on the Time (s) column. Data collection. EDA was recorded with a BITalino PsychoBIT unit at 1000 Hz via electrodes on the palm of the non-dominant hand, and is provided in microsiemens (µS). File structure. Raw EDA - 65 students: folder containing 65 files, one per anonymized participant, each with two columns: Time (s) + EDA (µS) grades.csv : links each participant ID to their mock exam grade (Mock_Exam_Grade). Only the mock exam grade is provided; real-exam scores are withheld to protect participant privacy. QuestionnairesResponses.csv : self-reported responses collected around the mock exam, all on a 1–5 Likert scale: Pre-exam: perceived stress (Stressed), tiredness (Tired), self-assessed ML knowledge (ML_Knowledge), and exam preparation (Well_Revised). And post-exam: self-rated performance expectation (Success1-5) Important note on scope. No real-exam grades are included, only mock exam grades tied to the recorded session. This makes the dataset suitable for studies relating EDA to performance on the same assessment being recorded, as well as for standalone signal processing, stress/cognitive workload detection, affective computing, and research combining physiological and self-reported measures of academic stress. Code. Analysis code accompanying this dataset is available on GitHub: [link TBD]