The Ninth Annual MLSP Competition: First place

Author

Fodor, Gabor

Author_Institution

Budapest Univ. of Technol. & Econ., Budapest, Hungary

fYear

2013

fDate

22-25 Sept. 2013

Firstpage

1

Lastpage

2

Abstract

The goal of the 2013 MLSP Competition is to predict the set of bird species present in audio recordings, collected in field conditions. Real-world audio data presents special difficulties such as simultaneously vocalizing birds, other animal sounds, and background noise. Although the task can be considered as a multi-instance multi-label learning problem, I propose a Binary Relevance approach with Random Forest. The proposed solution achieves 0.956 AUC and ranks 1st place on the Kaggle private leaderboard.

Keywords

audio recording; decision trees; feature extraction; image matching; image segmentation; learning (artificial intelligence); Kaggle private leader board; animal sounds; audio recordings; background noise; binary relevance approach; bird species; feature extraction; multiinstance multilabel learning problem; ninth annual MLSP competition; random forest; real-world audio data; template matching; unsupervised image segmentation method; Audio recording; Birds; Feature extraction; Image segmentation; Spectrogram; Whales; random forest; spectrogram; template matching;

fLanguage

English

Publisher

ieee

Conference_Titel

Machine Learning for Signal Processing (MLSP), 2013 IEEE International Workshop on

Conference_Location

Southampton

ISSN

1551-2541

Type

conf

DOI

10.1109/MLSP.2013.6661932

Filename

6661932