Enhancing Accuracy and Performance in Music Mood Classification through Fine-Tuned Machine Learning Methods

Abstract

Putting emotional labels on music, or "music mood classification," is important for use in recommendation systems and music therapy. Using fine-tuned machine learning methods, this study aims to improve the accuracy and performance of classification. We used a large dataset with names for different types of music and moods to make sure that the model training was strong. Advanced feature extraction methods picked up both the traits of the audio stream and the lyrics. For audio features, color features, spectral contrast, and mel-frequency cepstral coefficients (MFCCs) were recovered. For poetry analysis, TF-IDF and word embeddings were used, along with natural language processing (NLP) methods. Logistic Regression, SGD Classifier, Gaussian Naive Bayes, Decision Tree, Random Forest, XGB Classifier, SVM Linear, and K-Nearest Neighbors (KNN) were some of the machine learning classification methods we used. Random Forest, XGB Classifier, and SVM Linear all did better than the others. We used grid search and random search to fine-tune the hyperparameters of these top-performing models in order to make them even better. Cross-validation made sure that the models were stable and could be used in other situations. Our results show that the highly tuned Random Forest, XGB, and SVM models greatly improved the accuracy of classification, with the XGB Classifier performing the best. This study adds to music information retrieval by creating a useful method for mood classification that can be used in real-life situations to improve user experiences and create more personalized music services.