A multimodal dataset for emotional transition analysis in virtual reality

Emotion recognition from physiological signals typically treats emotions as discrete, static states rather than dynamic processes, limiting real-world affective computing applications. This dataset contains multimodal physiological recordings from 28 participants experiencing systematically designed emotional transitions in virtual reality. Participants viewed emotion-eliciting video stimuli across three emotional quadrants with transition periods between stimuli. Four physiological modalities were recorded: EEG (7 channels, 300 Hz), ECG (4 leads, 512 Hz), EMG (2 channels, 512 Hz), and GSR (3 channels, 10 Hz). The protocol employed a balanced incomplete block design across six emotional sequences. Statistical validation shows quadrant differentiation with 70% physiologically validated and 85% self-reported emotion induction success rates on average. Individual journey analysis indicates that participants traversed between 8.84% and 58.39% of the theoretical maximum cumulative distance on the valence-arousal plane, reflecting substantial individual differences in emotional responsivity. The dataset comprises 1.84 GB of original XDF recordings, 238 video-aligned physiological segments, and self-assessment ratings. This resource enables research in dynamic emotion recognition, and individual differences in responsivity during controlled emotional transitions.