Abstract:
The fault identification or prediction of faulty program’s essential but expensive process to maintain the quality and reliability of the software. Deep learning based fault prediction has been intensively studied recently and proposed several automated ways to test the program which help the developers to reduce the cost of manual labor in the process. In prior studies there are several traditional Learning to rank techniques which helps the developers to diagnose the fault locations like suspiciousness values. As the dimensions of features are increasing day by day and it needs some advance fault prediction techniques. Also in SBFL (Spectrum based fault localization) it reliesion the pass/fail results and coverage of the testicases. In this work we proposed deep learning approach for fault prediction which specifically deals with the classes by extracting the class level features from the program by combining the suits of class level metrics (CK, MOOD and Martin) and other source code metrics. A Public Unified Bug Dataset are used to get the class level metrics and then used their values as feature in deep learning algorithm. Deep Learning technique is applied to several datasets which are available and get the best results from all that datasets. The best combination of source code metrics is evaluated through the deep learning model prediction. To validate the results we use the same combination of class level metrics and evaluate the predicted result to the acutual results to validate the accuracy of this approach. Our model acheive the 92% of the accuracy from the Github bug dataset of 14 projects and compare our methodology with other algorithms.
