Good solutions to classification problems can have a significant impact on organization revenue and profitability. Typical high-impact classification problems in organizational decision-making, are associated with customer risk management (Bose & Chen, 2009) and organization default risk management, among others. A sub-set of the classification problem is the asymmetric misclassification cost classification problem, where the risk of misclassification is not symmetrical. Bult and Wittink (1996) identified three different kinds of misclassification risk, and these three different risks were: symmetric risk where all misclassification costs were equal, asymmetric homogeneous risk where misclassification costs were not equal but were constant, and asymmetric non-homogeneous risk where misclassification costs were not equal but were variable for each case. Zhao, Zhao, and Song (2009) have observed similar asymmetric risk in credit card markets. Data Envelopment Analysis (DEA) models for classification problems were first used in the 1990s (Troutt, Rai, & Zhang, 1994). Currently, there are a lot of DEA models available for a variety of business analytics tasks, such as solving inverse classification problems (Pendharkar, 2002), data preprocessing (Pendharkar, 2005), fuzzy classification (Pendharkar, 2012), interactive classification (Pendharkar & Troutt, 2014), cluster analysis (Toloo, Saen, & Azadi, 2015), feature selection (Zhang et al., 2015), and dimensionality reduction (Pendharkar & Troutt, 2011). The DEA models can also be incorporated into radial basis neural networks to solving non-linearly separable classification problems that may contain inputs with negative values (Pendharkar, 2011a). Most of the DEA based classification models use a dual variant of the variable returns-to-scale BCC model (Banker, Charnes, & Cooper, 1984) that provides non-linear (piecewise linear) classification decision boundaries. For two-class problems, a separate model is solved for each class, and two decision boundaries are obtained. When a problem is linearly inseparable, there are overlapping examples belonging to two different classes that appear between the two decision boundaries. However, when a classification problem is linearly separable, there are no such overlapping examples between the decision boundaries (see Fig. 1). For linearly separable problems, traditional margin maximizing support vector machines (SVMs) may be the best classifiers, and DEA models should not be used. DEA models are primarily suitable for linearly inseparable problems. These models may also be applied to nonlinearly separable problems when combined with neural networks as hybrid models (Pendharkar, 2011a). The primary contribution of this paper is to design and use DEA models for probabilistic classification. To our knowledge, probabilistic classification DEA models have never been used for such tasks.