 |
    |
Predicting Shellfish Farm Closures with Class Balancing Methods
Claire D’Este1, Ashfaqur Rahman1, and Alison Turnbull2
1Intelligent Sensing and Systems Laboratory and Food Future Flagship, CSIRO, Castray Esplanade, Hobart, Australia 7000
claire.deste@csiro.au
2Department of Health and Human Services, GPO Box 125, Hobart, Australia 7001
Abstract. Real-time environmental monitoring can provide vital situational awareness for effective management of natural resources. Effective operation of Shellfish farms depends on environmental conditions. In this paper we propose a supervised learning approach to predict the farm closures. This is a binary classification problem where farm closure is a function of environmental variables. A problem with this classification approach is that farm closure events occur with small frequency leading to class imbalance problem. Straightforward learning techniques tend to favour the majority class; in this case continually predicting no event. We present a new ensemble class balancing algorithm based on random undersampling to resolve this problem. Experimental results show that the class balancing ensemble performs better than individual and other state of art ensemble classifiers. We have also obtained an understanding of the importance of relevant environmental variables for shellfish farm closure. We have utilized feature ranking algorithms in this regard.
LNAI 7691, p. 39 ff.
lncs@springer.com
© Springer-Verlag Berlin Heidelberg 2012