Song-based Classification techniques for Endangered Bird Conservation.

Abstract The work presented in this paper is part ofa global framework which long term goal isto design a wireless sensor network able tosupport the observation of a population of en-dangered birds. We present the rst stage forwhich we have conducted a knowledge discov-ery approach on a sample of acoustical data.We use MFCC features extracted from birdsongs and we exploit two knowledge discoverytechniques. One relies on clustering-basedapproaches and highlights the homogeneityin the songs of the species. The other oneis based on predictive modeling and demon-strates the good performances of various ma-chine learning techniques for the identi ca-tion process. The knowledge elicited providespromising results to consider a widespreadstudy and to elicit guidelines for designinga rst version of the automatic approach fordata collection based on acoustic sensors. 1. Introduction In last decades, due to the exponential growth of globalcommercial and industrial activities, numerous scien- ICML 2013 Workshop on Machine Learning for Bioacous-tics, Atlanta, Georgia, USA, 2013. Copyright 2013 by theauthor(s).