Research and Development of Software Defect Prediction Model Based on Deep Learning

Le résumé fourni par la source

Most of the early software applications were developed for scientific computation and data processing. The software industry has experienced significant growth over the last few decades, mainly due to technological advancements. As a result, software has increasingly become an important part of today's world. With the increasing pervasiveness of software in everyday life, software engineers are required to meet strict criteria to develop trustworthy software. This has led to the increased use of software testing frameworks. Software testing involves a thorough assessment of the functionality of software systems to identify bugs. The automation of software testing is considered a viable solution to counter the complexity and cost involved in most testing activities. An ever-increasing proportion of software engineering activities, particularly those related to software testing, is being automated using machine learning. Conventional software defect prediction methods have relied heavily on traditional source code metrics like code complexity and lines of code. However, these methods have not been able to capture the semantics present in source code. This paper proposes a novel fine-tuned convolutional neural network model with a focus on correlation clustering and testing metrics. The proposed approach seeks to locate areas in the source code where defects, errors, or flaws are likely to be present. The Abstract Syntax Tree tokens are derived from the source code and used as feature vectors for testing metrics. This feature vector is then fed into a CNN. To improve the accuracy of defect prediction, the CNN model is further optimized by hyperparameter tuning.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.