Prediction of software defects by knowledge graph and genetic algorithm

Software defect detection is one of the biggest software development challenges and accounts for the largest budget in the software development process. One of the effective activities for software development and increasing its reliability is to predict software defects before reaching the test stage, which helps to save time in the production, maintenance and cost process. This research aims to present a software defect prediction method based on knowledge graphs and automated machine learning. We use knowledge acquisition, knowledge fusion, knowledge storage and knowledge calculation and other knowledge map construction technology research, to realize the knowledge map recommends high-quality software defect prediction models as the hot-start input conditions for automatic search. The empirical study uses NASA's open-source dataset experimental objects and six performance evaluation indicators include Precision, Recall, PRC (Precision Recall Characteristic), ROC (Receiver Operating Characteristic), F-Measure, MCC (Matthews Correlation Coefficient). The experimental results show that the proposed model performs better than the traditional classic software defect prediction model recommended by the knowledge map in terms of different datasets and evaluation indicators