Abstract: To solve the problems that the existing software defect assessment methods do not generate processing priority according to the dichotomous classification results, and the sample-based class imbalanced processing methods are prone to cause important data loss, a software defect assessment method with classification cost is proposed. First, preprocessed samples are obtained by processing multi-source samples through four phases: sample noise processing, sample distribution normalization, sample balance reconstruction and feature dimensionality reduction. Then, the evaluation indexes and index weights on multiple classifiers are calculated to obtain the classification cost, and the heterogeneous weighted voting model is constructed through the classification cost. Finally, the heterogeneous weighted voting model is used to calculate the defective probability values and assess the levels of the defective samples. The experimental results show that the defective probability values and defect levels obtained by the proposed method are more consistent with the actual situation of the real samples, which can accurately reflect the defect levels of the code modules.

Key words: defect assessment, class imbalance , classification cost , heterogeneous weighted voting , defect level