Cost-Sensitive Neural Network with ROC-Based Moving Threshold for Imbalanced Classification
- 1.3k Downloads
Pattern classification algorithms usually assume, that the distribution of examples in classes is roughly balanced. However, in many cases one of the classes is dominant in comparison with others. Here, the classifier will become biased towards the majority class. This scenario is known as imbalanced classification. As the minority class is usually the one more valuable, we need to counter the imbalance effect by using one of several dedicated techniques. Cost-sensitive methods assume a penalty factor for misclassifying the minority objects. This way, by assuming a higher cost to minority objects we boost their importance for the classification process. In this paper, we propose a model of cost-sensitive neural network with moving threshold. It relies on scaling the output of the classifier with a given cost function. This way, we adjust our support functions towards the minority class. We propose a novel method for automatically determining the cost, based on the Receiver Operating Characteristic (ROC) curve analysis. It allows us to select the most efficient cost factor for a given dataset. Experimental comparison with state-of-the-art methods for imbalanced classification and backed-up by a statistical analysis prove the effectiveness of our proposal.
KeywordsMachine learning Neural networks Imbalanced classification Cost-sensitive Moving threshold
This work was supported by the Polish National Science Center under the grant no. DEC-2013/09/B/ST6/02264.
- 4.Flach, P.A.: The geometry of ROC space: understanding machine learning metrics through ROC isometrics. In: Proceedings of the Twentieth International Conference on Machine Learning, ICML 2003, 21–24 August 2003, Washington, DC, USA, pp. 194–201 (2003)Google Scholar
- 9.Maloof, M.A.: Learning when data sets are imbalanced and when costs are unequal and unknown. In: ICML-2003 Workshop on Learning from Imbalanced Data Sets II (2003)Google Scholar
- 10.Provost, F.: Machine learning from imbalanced data sets 101. In: Proceedings of the AAAI 2000 Workshop on Imbalanced Data Sets, pp. 1–3 (2000)Google Scholar
- 11.Riedmiller, M., Braun, H.: A direct adaptive method for faster backpropagation learning: the RPROP algorithm. In: 1993 IEEE International Conference on Neural Networks, pp. 586–591 (1993)Google Scholar