Assessing performance of different classifiers and selecting the best one is one of the most important tasks in classification problem. The assessment of classifiers becomes more complex when dataset is imbalanced because most of the frequently used performance metrics can be misleading. Many real world classification problems such as fraud detection, churn prediction, medical diagnosis, and cyber-security suffer from the problem of imbalanced datasets. Therefore, in all such classification tasks it is very important to select the best classifier very carefully. In this study we propose new weighted TPR-TNR measure to assess performance of the classifiers. The proposed measure takes into consideration imbalance ratio of the dataset and assign different weights to the TPR and TNR to assess classifiers performance. We have used five different datasets to assess performance of twelve different classifiers using weighted TPR-TNR measure and compared it with the existing measures. The experimental results show that the weighted TPR-TNR measure is more suitable to assess performance of the classifiers when dataset is imbalanced.

评估不同分类器的性能并选择最佳分类器是分类问题中最重要的任务之一。当数据集不平衡时，分类器的评估变得更加复杂，因为大多数常用的性能指标可能会产生误导。诸如欺诈检测，客户流失预测，医疗诊断和网络安全之类的许多现实世界中的分类问题都受到数据集不平衡的困扰。因此，在所有此类分类任务中，非常仔细地选择最佳分类器非常重要。在这项研究中，我们提出了新的加权TPR-TNR度量来评估分类器的性能。拟议的措施考虑了数据集的不平衡率，并为TPR和TNR分配了不同的权重，以评估分类器的性能。我们使用了五个不同的数据集，使用加权TPR-TNR度量来评估十二个不同分类器的性能，并将其与现有度量进行比较。实验结果表明，当数据集不平衡时，加权TPR-TNR度量更适合评估分类器的性能。