A Deep Learning Based Model for Defect Prediction in Intra-Project Software

Testing of software ensures the supply of meaningful software and hence prediction of defects in producing high quality software has become an inevitable one. Software defect prediction's main aim is to find out various bugs present in software and focus on testing efforts. Many of existing software defect prediction frameworks are much simple, making it difficult for developers to get detailed reference information. Nowadays, many deep learning models, like the Radial Base Functional Neural Network(RBF) and the Convolutional Neural Network (CNN), are applied to features which are created automatically from deep learning models and abstract syntax trees (AS Ts) to aid in the improved performance of predicting defects. But the results generated using RBF and CNN algorithms are not able to provide much accuracy due to its restricted size of dataset and improper baseline selections. To resolve these state-of-the-art problems, we have constructed a dataset taken from various defect datasets namely the Kamei Dataset, NASA Dataset and the PROMISE Source Code (PSC) dataset. In this research, the dataset is named as Combination Defect Analysis Dataset (CDA). Then, an Enhanced Convolutional Neural Network (ECNN) model is proposed for predicting defects in Intra-Project software (IPDP) and associated results to different models. Experimental results implied that Enhanced CNN(ECNN) model is efficient compared to the other associated models, along with it outclassing the other machine learning models suggested for IPDP.