Applying Feature Selection Combination in Audios of Whale for Improving Classification

Abstract

Audio signal processing has been under investigation for the last decades. The majority of the works found in literature focus on signal analysis and classification. Most of them integrate Machine Learning (ML) algorithms with the audio signal processing techniques. As the performance of any ML algorithm depends on the features of a dataset used for training and testing purposes, using a dataset derived from the extraction of features from an audio is not trivial due to the fact that the correct combination of extraction techniques with the selection of the most relevant attributes needs to take place. In this sense, this paper proposes an empirical analysis on different audio extraction techniques combined with feature selection for improving Whale audio classification. Usually, the application of audio extraction techniques results in poor classification performance. However, the combination of feature selection can achieve better results. The experimental results have been promising, indicating that the idea of combining different audio extraction techniques with feature selection can improve the performance of ML classification algorithms over whales’ audios by 22 percentage points.