入侵检测分类引擎预测精度度量方法

A New Measurement of Prediction Accuracy for Classification Engines in Intrusion Detection

在综合考虑入侵检测分类引擎对异常和正常记录整体预测能力的基础上，提出了一种与测试和目标审计记录集中的异常记录分布无关的预测精度度量方法。该度量方法消除了当前评估规范存在的缺陷，能够确保预测精度就是实际入侵检测时的泛化精度。

A new measurement method used for predicting accuracy of classification engines in the intrusion detection system is proposed on the basis of integrated consideration to the predictive ability of classification engines to normal and abnormal records. The proposed measurement is independent of the distribution of abnormal records in test and real audit sets and completely eliminates deficiencies of the current criterions. The predictive accuracy is assuredly the generalization accuracy in the actual intrusion detection.