Feature Selection in Music Data with Meta-Heuristic Methods

Abstract

In today's world, the rapid increase in multimedia content production has made accessing valuable information more challenging. Data mining has become critical to facilitate access to meaningful data, and an important step in this process is reducing the size of the data. Feature selection reduces the data size by eliminating irrelevant, noisy, or missing data from the dataset, allowing the methods used in data analysis to operate faster and more efficiently. In this study, feature selection was performed using nature-inspired metaheuristic algorithms. The selected features were analyzed by classifying them using machine learning algorithms and artificial neural networks. Improvements made on the music dataset increased track popularity classification performance by 3,2%, achieving an accuracy of 88%. The methods used were presented comparatively, and the findings were evaluated.