基于知识图谱和自动机器学习的软件缺陷预测

不稳定和召回率低效的软件缺陷预测模型难以在行业领域应用，为解决稳定和高效各项性能评价指标的软件缺陷预测模型在工程实践应用的问题，提出了一种基于知识图谱和自动化机器学习的软件缺陷预测方法AutoKGGAS，首先获取软件缺陷预测模型数据，对知识建模、知识获取、知识融合、知识储存与知识计算等知识图谱构建技术研究，实现知识图谱推荐优质软件缺陷预测模型作为自动化搜索的热启动输入条件，根据不同的软件缺陷预测评价指标，优化不同最佳的模型结构.其次实证研究采用NASA开源数据集实验对象和六种性能评价指标，实验结果表明， AutoKGGAS自动化软件缺陷预测模型在不同数据集不同评价指标方面，性能优于知识图谱推荐的传统经典软件缺陷预测模型.自动化软件缺陷预测模型为航天软件缺陷预测辅助代码审查测试提供了原型，在工程实践应用方面具有重要的意义.

Software Defect Prediction Based on Knowledge Graphs and Automatic Machine Learning

The software defect prediction model characterized by instability and low recall rate is difficult to apply in the industry field. To solve the problem of the software defect prediction model with stable and efficient performance evaluation indicators in the engineering practice, a software defect prediction method Automated knowledge graphs genetic algorithm stacking (AutoKGGAS) is proposed based on the automated knowledge graphs assisted machine learning, which obtains the software defect prediction model data for the research on knowledge graph construction technologies (such as knowledge modeling, knowledge acquisition, knowledge fusion, knowledge storage and knowledge calculation), take the high quality software defect prediction model recommended by the knowledge graphs as the hot startup input of automatic search. According to different software defect prediction evaluation indicators, different optimal stacking model structures are optimized. On the other hand, the empirical research uses NASA open source dataset experimental object and six performance evaluation indicators. The experimental results show that the AutoKGGAS automated software defect prediction model is superior to the traditional classic software defect prediction model recommended by the knowledge graphs in different evaluation indicators of different datasets. The automated software defect prediction model provides a prototype for the aerospace software defect prediction to assist the code review test, which is of great significance in engineering practices.