Semi-supervised learning software defect prediction method based on spectral clustering

Abstract

The invention discloses a semi-supervised learning software defect prediction method based on spectral clustering, and the method comprises the following steps: 1), obtaining original data, carrying out the data preprocessing, and obtaining a processed feature matrix; 2) judging whether the feature matrix has a label or not: clustering the data without the label through spectral clustering; performing label operation on the obtained cluster through a heuristic rule of software defect prediction to obtain a pseudo label, and then turning to the step 3); for the labeled data, directly going to the step 3); 3) calculating a feature deviation score according to the data distribution and performing feature selection, wherein the weight of the original label data is greater than the weight of the pseudo label data; and 4) performing clustering and labeling operation again according to the new feature matrix to obtain a prediction result. According to the method, the influence of irrelevant and redundant characteristics on the model result is reduced, the information of the original label data of the project is utilized, the accuracy of the software defect prediction result can be effectively improved, and the applicability of the model is improved.

Description

technical field [0001] The invention relates to the field of software defect prediction, in particular to a software defect prediction method based on spectral clustering semi-supervised learning. Background technique [0002] Software defect prediction is a process of predicting whether a software entity is defective or not. As the scale of contemporary software continues to expand, software defect prediction, as a technology that can help reduce the burden of software testers and optimize the allocation of developers and testers, has received more and more attention. Practice has shown that the cost of finding and fixing defects after development is completed is much higher than the cost of finding and fixing defects during development. Therefore, it is very important to introduce software defect prediction in the early stage of software life cycle. But at present in the industry, the application of software defect prediction is still less. This is mainly because the cu...

AI Technical Summary

Problems solved by technology

However, the data owned by the actual project is often a collection of a small number of labeled data and most of the unlabeled data. The above two schemes assume that the target project does not have any historical label data, which loses part of the available known information and reduces the model. performance

[0004] In addition, the main focus of existing research is to construct a usable model without focusing on feature selection

However, the use of a large number of irrelevant or redundant features not only wastes computing resources, but also degrades the performance of the prediction model

Images