Improved Automatic Bird Identification through Decision Tree based Feature Selection and Bagging

This paper presents a machine learning technique for bird species identification at large scale. It automatically identifies about a thousand different species in a large number of audio recordings and provides the basis for the winning solution to the LifeCLEF 2015 Bird Identification Task. To process the very large amounts of audio data and to achieve similar good results compared to previous identification challenges new methods e.g. downsampling of spectrogram images for faster feature extraction, advanced feature selection via decision tree based feature ranking and bootstrap aggregating using averaging and blending were tested and evaluated.