LLaMAC: low-cost biosignal sensor based large multimodal dataset for affective computing

The LLaMAC dataset was developed to predict the success of audio-visual media via emotion prediction. It was created using low-cost biosignal sensors, with emotional questionnaires in both continuous (valence, arousal, dominance) and discrete domains (emotion type and intensity: neutral, fun, sadness, anger, fear), and included over 100 participants. Questionnaires on liking and familiarity were also collected. The dataset contains five biosignals-EEG, GSR, PPG, SKT, and RESP-and seven questionnaire measures. Biosignals were validated using statistical metrics and signal-to-noise ratios, while questionnaire responses were assessed with scatter plots and statistical analyses. Emotion classification was performed using a Light Gradient Boosting Machine (LightGBM). The dataset enables biosignal-based prediction of emotions and liking, correlation analysis between continuous and discrete emotions, and investigation of biosignal differences related to familiarity, which can further inform emotion and liking predictions.