Enhancing Music Emotion Classification with Lyrics and Audio Features

The role of automatic generated playlists and recommendations in music streaming services has played a significant role in driving the steady growth of the music industry since 2015. With the evolving landscape of music, Music Emotion Recognition (MER) has gained importance in this era. Contemporary solutions today focus solely on four emotions and popular music genres, neglecting emerging styles and artistes where their music evokes a danceable rhythm, but lyrics convey negativity. This study considers both lyrics and audio features for the classification of eight emotions with the aim to identify key features influencing music emotion, develop a classification model and evaluate its performance. A dataset was created by expanding a base dataset with additional lyrics and audio features from Genius and Spotify API respectively. This study employs established algorithms like K-Nearest Neighbours (KNN), Decision Trees (DT), and Random Forests (RF), with Random Forests emerging with a notable 73% accuracy. A web application named Emotion was developed, allowing data scientists and music researchers to explore the dataset and access details of this study. This study's impact extends to various applications, including mental health and media, enabling tailored music-based interventions for emotional well-being, and improving music selection in media and entertainment.