Feature Selection for Vocal Segmentation Using Social Emotional Optimization Algorithm
Feature selection is an important task in many applications of pattern recognition and machine learning areas. It involves in reducing the number of features required in describing the large set of data. Many practical problems often have a large number o
- PDF / 1,427,104 Bytes
- 23 Pages / 439.37 x 666.142 pts Page_size
- 3 Downloads / 257 Views
Abstract Feature selection is an important task in many applications of pattern recognition and machine learning areas. It involves in reducing the number of features required in describing the large set of data. Many practical problems often have a large number of features in the data sets, but not all of them are useful for the pattern recognition algorithms such as classification. Irrelevant, and redundant features may even reduce the performance. Feature selection aims to choose a small set of relevant features to achieve the same or even better performance of the classification algorithm. However, it is a challenging task to choose the best subset of features due to the large search space. It is considered as an optimization problem which tries to select the best subset of features from the complex search space that improves the performance of the algorithm. A binary version of the Social Emotional Optimization Algorithm (BSEOA) is proposed for feature selection in classification problems. The algorithm is tested on benchmark datasets for the classification using the Support Vector Machine (SVM) as the classifier. Also, the algorithm is used for selecting the features which can be used for the vocal segmentation of the collected songs. The vocal segmentation problem is considered as the classification of the vocal and nonvocal parts of the song. The experimental results show that the proposed binary SEOA is efficient in improving the classification accuracy by selecting an optimum set of features.
1 Introduction In pattern recognition and machine learning, most of the real-world classification problems involve a large number of features to describe the data. These features P. Rajasekharreddy (B) · E. S. Gopi Department of Electronics and Communication Engineering, National Institute of Technology, Tiruchirappalli, Tamil Nadu, India e-mail: [email protected] E. S. Gopi e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2019 A. J. Kulkarni et al. (eds.), Socio-cultural Inspired Metaheuristics, Studies in Computational Intelligence 828, https://doi.org/10.1007/978-981-13-6569-0_4
69
70
P. Rajasekharreddy and E. S. Gopi
include relevant, irrelevant, and redundant features. However, irrelevant and redundant features are not useful and they may even reduce the performance of the algorithm due to the large search space known as “Curse of dimensionality” [1]. In real-time problems, the cost (such as computational cost, time, etc.,) involved in obtaining some features is high. If the features that are really helpful for the pattern recognition algorithms are known then the cost involved in collecting the redundant and irrelevant features can be reduced. Feature selection can address this problem by selecting a small number of features so as to retain or improve the performance of the classification algorithm. By removing or reducing the irrelevant and redundant features, feature selection can reduce the dimensionality of the data, speed up the training of the algorithm, and/or improve the performanc
Data Loading...
