A new accuracy evaluation criterion for binary classifiers

A new evaluation criterion used for predicting accuracy of binary classifiers for identifying high risk components in software engineering is proposed on the basis of analysis of the presented measure criterions so as to achieve the highest possible prediction accuracy. The proposed evaluation criterion completely eliminates deficiencies of the current criterions, and is independent of the proportion of high risk components in the component set. Many different examples make sure that results of evaluation are assuredly the results of prediction.