Software Defect Prediction by Logistic Regression with Gradient Descent Cost Computation

Ever since software development, bug free software is major challenge to all software industry. Both software developers and testers require ongoing enhancements to their development and testing processes to deliver high-quality software. Early detection and assessment of software project failure can aid in improving project outcomes and guiding them towards success. During the software development process, several risks could have an impact on the project and perhaps cause it to fail. Making a distinction between classes with and without defects from continuous defect counts is common procedure, when developing defect prediction model and apply it as a class attribute. Regression models could be used to create defect classifiers by categorizing the projected defect counts into classes that are defective and non-defective. In this study we developed model for bug prediction by computing Gradient Descent and cost value for Logistic regression. We have performed our experiment on CM1 defect dataset and result shows that logistic regression gives good accuracy for defect prediction when used with Gradient Descent and cost function. We therefore recommend that regression-based classifiers should be used for further defect classification research, especially in cases when the modelled dataset's defective ratio is miserable.