Data for Subject-Adaptive Multimodal Framework
Description
This dataset contains time-synchronized facial geometric features, vehicle telemetry, and contact-based autonomic physiological signals recorded from 31 licensed drivers during a stress-inducing driving-simulator protocol. It is the dataset used to train and evaluate the Random Forest and dual-stream CNN-LSTM models reported in [PAPER CITATION], where facial and driving signals are used to infer binary sympathetic (electrodermal) and parasympathetic (heart rate variability) states under four personalization strategies. Participants and protocol. Thirty-one licensed drivers (21 male, 10 female; mean age 27.7 years; mean driving experience 6.6 years; minimum two years of licensed driving) completed a Unity-based high-fidelity driving-simulator course. Participants who wore glasses were excluded to preserve facial-landmark precision. The course comprised nine distinct stress-inducing scenarios delivered across 16 sequential events, including environmental stressors (heavy rain, sun glare) and dynamic traffic conflicts (abrupt lane changes by adjacent vehicles, roadway obstacles, slow-moving vehicles blocking the path). Each event lasted 20–90 s and was separated by neutral driving intervals to permit partial physiological stabilization. The session structure was a three-minute familiarization drive followed by two full driving sessions separated by a five-minute rest. Data were collected between January and May 2025. Contents. A single CSV file with 299,716 rows and 49 columns. Each row is one synchronized sample; the three data streams were captured at 5 Hz. Records are grouped by participant (groups, 31 unique IDs) and by driving session (session_id, 62 sessions). Per-participant record counts range from 7,869 to 11,084 (median 9,721). Columns fall into six groups: Timing and identifiers — timestamp, date, time, time(ms), datetime, delta, groups (participant ID), session_id, new_session Electrodermal activity — GSR (skin conductance, µS), GSR_z (within-participant z-score), SCR, SCR_z (phasic skin conductance response) Cardiac — PPG (raw photoplethysmography), IBI (inter-beat interval, ms), HR, HRV, SDNN, RMSSD, pNN50 Facial geometry (18 features) — Leye_X, Leye_Y, Reye_X, Reye_Y (pupil coordinates); Leye_open, Reye_open, Left_eye_h, Right_eye_h (eye openness and height); Left_brow_h1, Right_brow_h1, Left_brow_h2, Right_brow_h2, Between_brows (brow position and spacing); Left_chin_length, Right_chin_length (jawline contour); Mouth_width, Mouth_inner_height, Mouth_outer_height (mouth geometry) Facial expression classification — Angry, Disgust, Fear, Happy, Neutral, Sad, Surprise (class probabilities) and Emotion (argmax label) Vehicle telemetry — speed (km/h), brake (pedal pressure), handle (steering angle)