Binary Classification, Probabilities, and Evaluating Classification Performance
Johannes Ledolter · 2013
Many decision problems can be reduced to a binary classification. In order to make a decision on a new case, it is necessary to estimate its success probability. The two ways to go wrong in a binary problem are false positive error and false negative error. The cutoff on the probability of success determines how items are classified. The classification rule depends on the class probability and on the probability cutoff. The two classification rules are sensitivity and specificity. The cutoff on the probability affects the trade-off between sensitivity and specificity, and the receiver operating characteristic (ROC) curve illustrates this graphically. The chapter illustrates an example: German credit data.