Music Mood Prediction and Playlist Recommendation based on Facial Expressions

This research paper presents a comprehensive approach to music recommendation, combining facial emotion recognition through image processing with the analysis of music attributes. The project comprises 3 distinct stages, culminating in a novel personalized music recommendation system. In the initial stage, advanced image processing techniques and the DeepFace library are utilized to accurately detect and classify facial expressions, enabling the identification of emotions such as happy, sad, energetic, and calm. The second stage involves constructing a robust predictive model that correlates music attributes with specific mood states based on a dataset, providing valuable insights into the relationship between music attributes and emotions. 10 algorithms are trained and tested and the best fit among them is recognized. Predictions are made again on the 2nd dataset which will be later used for song recommendation. Building upon these insights, the third stage combines results of facial emotion recognition and mood prediction to recommend the top 40 songs from a dataset that best aligns with the user's current emotional state. This interdisciplinary research project bridges machine learning and music analysis to create an innovative music recommendation system. The evaluation of this system through experiments and user studies demonstrates its potential to enhance music discovery and user satisfaction, offering a novel approach to tailoring music recommendations based on the user's detected emotional disposition.