Wrapper Feature Subset Selection for Feature Extraction of Bonang Barung Single Tone Convertion Into Numeric Notation

Abstract

Several researches have been done to study the characteristics of the bonang barung, one of Javanese Gamelan music instrument. One of them is convertion of bonang barung single tone to numeric notation using Harmonic FFT as feature extraction and Backpropagation Artificial Neural Network (ANN) for classification. The tone detection accuracy result from previous research is 70,74%. In this research we try to improve the detection result by searching the dominant features using Wrapper Feature Subset Selection (WFSS). Sequential forward selection (SFS) and sequential backward selection (SBS) are used as searching algorithm. The input of the system is a song recorded from a bonang barung then the detected tone is converted into numeric notation. From the experiment, WFSS-SFS produced 6 features with 86,4% accuracy while WFSS-SBS give a better result, it produced 13 features with 92,9% accuracy of tone detection.