Soft Computing Techniques for Software Fault Prediction: Comparative Analysis
Abstract - This research paper compares the performance of software fault prediction model developed using Bayesian net, FIS and ANFIS with specific focus on process level software metrics during software development. The developed models used for comparison are designed using software metrics as input data to compute the fault density for each basic phase of SDLC. The comparative study is built on validation of the prediction accuracy for various soft-computing modeling techniques. Furthermore, the comparative study is performed on the data set from the PROMISE repository, and the outcomes of various soft computing methodology based models has been compared. The study shows that ANFIS based model for software fault prediction provides better accuracy as calculated from statistical analysis, MMRE (0.016461) and RMSE (0.0375) in contrast to other soft computing techniques based on a model on Bayesian net and FIS.
Keywords - Software development life cycle (SDLC), Software Fault Prediction (SFP), Soft Computing, Fuzzy Inference System (FIS), Neuro-Fuzzy System (NFS), Adaptive Neuro-Fuzzy Inference System (ANFIS), Software Fault, Artifical Neural Network (ANN).