The researches on feature selection play a very important role in the area of classification. In this paper, we introduce a heuristic wrapper method: Classification Contribution-Ratio Based Selection (CCRS). Using RBF neural network as a classifier, we did our experiments on a data set of 74 features extracted from 517 Chinese folk songs which come from 10 regions. The results show that Root Mean Square, Spectral Flux and Linear Prediction Coefficient are very effective for the classification of 10 kinds of Chinese folk songs. It works better when the number of features is reduced from 74 to 30 and the classification accuracy is improved from 39.74% (using the total 74 features) to 43.208% (using 30 optimal features). At last, we give an illustration of validity of the algorithm, and a comparison with the Fisher Criterion Method which shows the efficiency of CCRS.
Feature Selection for Automatic Classification of Chinese Folk Songs
2008 Congress on Image and Signal Processing ; 5 ; 441-446
2008-05-01
380324 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
A novel feature selection framework for automatic web page classification
British Library Online Contents | 2012
|Algorithm composition of Chinese folk music based on swarm intelligence
British Library Online Contents | 2017
|Feature Detection with Automatic Scale Selection
British Library Online Contents | 1998
|