Music Mood Classification using Python

Automatic recognition of emotions in musical audio has gained increasing attention in the field of music information retrieval (MIR) during the past few years. The development in the field has coincided with the need for managing large collections of digital audio for the public via web services such as Spotify and Last.fm. This is reflected, for example, in the increasing number of submitted systems in the annual Audio Music Mood Classification (AMC) contest part of the Music Information Retrieval Evaluation eXchange3 (MIREX). The substantial variance in the submitted systems in the contest and in the approaches in the MIR field in general indicates a lack of consensus concerning the choice of an underlying psychological model for emotions or moods in music and the establishment of precise machine learning conventions. Despite research in musicology studying the influence of specific cues in the musical structure on emotional expressions, there is no complete analytical consensus about the required “ingredients”—i.e., acoustical features extracted from music—to build the optimal models for emotion recognition, in the limits imposed by the subjectivity of [login to view URL] descriptors for each music track provided for this activity are the state-of-the-art descriptors used in this area (see table on next page for details). Your task is to determine which sets of features are most useful for learning the accurate classifiers. In particular your task isto explore different feature sets and machine learning algorithms with the aim to obtain the most accurate classifier. As part of your exploration you are supposed to get the corresponding accuracies for several algorithms (e.g. SVM, KNN, Decision Trees, Neural Networks, Ensemble, Naive Bayes), and applying training set, bootstrap, and cross-validation methods to get the training and testing accuracies. What kind of features (timbre, rhythm, pitch, dynamics, structure or harmony) are more relevant for predicting emotion in music in the provided data set?