In recent years, the increasing popularity of music streaming platforms has made it possible to listen a vast array of songs. Therefore, methods to recommend songs that match the preferences of various users have been attracting increasing attention. Traditional methods of music recommendation often rely on the user’s playback history and explicit feedback, which may not accurately reflect users’ actual preference. However, biological parameters such as photoplethysmogram (PPG) and galvanic skin response (GSR) reflect the electrical activity of the heart and the skin’s conductance, providing more accurate preference recognition. In this study, we used short music clips to assess individual preferences, utilizing sensors to record the associated physiological signals and compiling datasets for the experiment. During the post-processing stage, we extracted features from the physiological signals and built a classification model using Random Forest Classifier (RF) and Support-Vector Machine (SVM).