Data-driven cymbal bronze alloy identification via evolutionary machine learning with automatic feature selection

This paper aims to implement four machine learning models using Differential Evolution to tune internal parameters and for feature selection in a problem involving the classification of drum cymbals according to their bronze alloys via their sound. In order to conduct the experiments, 276 audios referring to 4 cymbals were captured at a recording studio with a controlled environment and conditions. Then, 18 temporal attributes were extracted from each audio file, aiming to retrieve information from them. The experimental results show that the Extreme Gradient Boosting model combined with Differential Evolution for parameter tuning showed consistent results in all performance metrics. Furthermore, when this evolutionary algorithm selects the attributes, a considerable increase in performance is obtained, reaching 98.90% average accuracy.