As mentioned in Section 2.1.1, model selection is an important part of the process of designing pattern recognition systems. In order to perform model selection, we need a way to compute and compare the accuracy of different classification algorithms. The accuracy of a classifier C is defined as (1 − Perr ), where Perr is the probability of error, i.e., the probability of a pattern x be misclassified by C. The accuracy can be directly estimated as the fraction of correctly classified patters. For example, assuming the classifier C has been trained using a training dataset D, given a labeled dataset of test examples T = {(z1 , L(z1 )), (z2 , L(z2 )), .., (zNt , L(zNt ))}, with T , D, the accuracy can be computed as
đang được dịch, vui lòng đợi..
