摘要：

Business operators and stakeholders often need to make decisions such as choosing between A and B, or between yes and no. These decisions include, but are not limited to, whether to invest in project A versus project B, or whether to continue running a company. These are often made by using a classification tool or a set of decision rules. For example, banks often use credit scoring systems to classify lending companies or individuals into a high or low risk of default, thus helping to decide whether to grant a loan. One important question businesses need to answer is how accurate the information based on these classification tools can help them make a correct decision, or how correctly they can be used to discriminate between two groups of subjects. In this chapter, we address this important issue by presenting accuracy parameters for assessing classification tools such as test modalities, scoring systems, and prediction models. Specifically, we introduce the receiver operating characteristics (ROC) curve as a statistical tool to evaluate these modalities. The ROC curve is widely used in business optimization analysis, health policy making, clinical studies, and health economics (Kampfrath & Levinson, 2013). In the Background section, we give updated examples of using the ROC related methods for assessing decision-makings based on our most current literature review. In the Main Focus section of this chapter, we provide mathematical definitions of the classification accuracy parameters, and describe the procedure to obtain an ROC curve. In addition, we present recent statistical developments in ROC curve methodologies and applications of ROC analysis in a diversity of research areas.

展开