To improve the prediction performance of the severity of software defects, considering the order between different labels, this paper proposed a defect severity prediction method ORESP based on ordinal regression. This method used the Spearman-based feature selection method to identify and remove redundant features in the data set and then used a neural network based on the proportional odds model to build prediction models. By comparing with the five classical classification methods, the proposed method ORESP could achieve better prediction performance under four different categories of module metrics. In terms of MZE(average zero-one error) measure, the performance of the proposed model can be improved by 10.3% at most; in terms of MAE(mean absolute error), the performance of the proposed model can be improved by 12.3% at most. In addition, using the feature selection method based on Spearman can effectively improve the prediction performance of ORESP.

[2021-06-05] Printed Article /doc/2020.07.0249/v2