A Software Defect Prediction Method Based on Program Semantic Feature Mining

As the size and complexity of software systems grow, knowing how to effectively judge whether there are defects in the programs has attracted extensive attention in research. However, current software defect prediction methods only extract semantic information at the syntactic level and lack features to mine defect manifestations at the semantic level of code, because defective software is incomplete or defective in semantic representation. Defective software exhibits incomplete or flawed semantic behavior. This paper proposes a software defect prediction method based on the program semantics feature mining (PSFM) method. Specifically, the semantic information is first extracted from the code grammatical structure information and code text information. Then, the defect feature is mined through the semantic information. Finally, software defects are predicted by using the mined defect features. The experimental results show that, compared with the existing software defect prediction methods, the method in this paper (PSFM method) obtained a higher F-measure value.