Music Mood Classification as a Subsystem of a Music Recommender System for Human Well-Being

This work introduces a system for the automatic classification of music mood em- ploying the Random Forest algorithm and audio feature extraction with the Librosa library. The suggested solution utilizes prominent audio descriptors, Mel-frequency cep- stral coefficients (MFCCs), and Mel spectrograms, in order to achieve the timbral and spectral qualities of music pieces. These audio features are obtained from a cleaned dataset of tagged audio files reflecting different mood labels. Visualization methods like Mel spectrogram and MFCC plots are utilized to investigate the feature space and facil- itate interpretability. A Random Forest classifier is trained using the features extracted to classify tracks into pre-defined mood labels. Experimental results prove the success of this approach in extracting emotional content from audio and obtaining competitive classification accuracy. The union of strong audio features and classical machine learning emphasizes a powerful yet light method for music mood analysis.