Optimization of General Statistical Accuracy Measures for Classification Based on Learning Vector Quantization.
Marika Kaden, Wieland Hermann, Thomas Villmann · 2014
Abstract. We propose a framework for classification learning based on generalized learning vector quantization using statistical quality measures as cost function. Statistical measures like the F-measure or the Matthews correlation coefficient reflect better the performance for two-class classification problems than the simple accuracy, in particular if the data classes are imbalanced. For this purpose, we introduce soft approximations of those quantities contained in the confusion matrix, which are the basis for the calculation of the quality measures. 1