Machine Learning and Deep Learning, emotions produced by music: A systematic review
The purpose of the study lies in the systematic review of research papers that focus on Machine Learning and Deep Learning for the prediction of music mood and the recognition of human emotions, for this purpose, data have been collected from documents published in the following online databases: IEEE Xplore, Scopus, ACM Digital Library, ScienceDirect, Mendeley and Hindawi. Aplicando los criterios de inclusión y exclusión, se logró compilar 44 artículos como fuentes primarias, considerando 3 elementos, técnicas, características más usadas y tipo de emociones predichas en las canciones y las métricas de evaluación siendo de base para dar respuesta a las preguntas de investigación. Thus, the results obtained show the most used techniques regarding the prediction of the mood of a song, being: Suport Vector Machine (SVM) = 14.71%, Convolutional Neural Network (CNN) = 11.76%, K Neighborhood Classification (KNN) = 7.35% and Logic Regression (LR) = 7.35%; Thus, the most used indicators to determine the mood of a song were Valencia (17.89%) and Arousal (15.79%); The main emotions were; Happy (24.55%) and Sad (22.73%); the most used metrics: Accuracy (42.65%), Accuracy (10.29%), Recall (8.82%), F1 score (7.35%) and Mean Absolute Error (5.88%).