Decoding emotions: unveiling the potential of facial landmarks

This study aimed to validate emotion decoding using a facial landmark-based prediction model trained on real-world data. We used facial landmark data and behavioral ratings measured while 29 participants (21 female, 22 male; age = 19–29, M = 22.37, SD = 2.25). viewed short video clips eliciting 10 different emotions including amusement, anger, awe, disgust, enthusiasm, fear, liking, sadness, surprise, and neutrality. Cross-participant classifications with support vector machine classifier were conducted to predict emotions across individuals. The results demonstrated promising predictive capabilities, achieving over 40% accuracy for behavioral ratings, facial landmarks, and their combination. Notably, classification based on combined features reached 75% accuracy, excelling in emotions like disgust, fear, and surprise. In summary, this study demonstrates the ability of facial landmark-based model to decode emotions accurately, particularly emphasizing negative and high-arousal emotions, and underscores the importance of supplementary tools in enhancing emotional response prediction.