Machine Learning Empowered Software Defect Prediction System

Production of high-quality software at lower cost has always been the main concern developers. However, due to exponential increases in size and complexity, development qualitative with costs is almost impossible. This issue can be resolved by identifying defects early stages lifecycle. As a significant amount resources are consumed testing activities, if only those modules shortlisted for that identified as defective, then overall reduced assurance high quality. An artificial neural network considered one extensively used machine-learning techniques predicting defect-prone modules. In this paper, cloud-based framework real-time software-defect prediction presented. proposed framework, empirical analysis performed compare performance four training algorithms back-propagation technique on prediction: Bayesian regularization (BR), Scaled Conjugate Gradient, Broyden–Fletcher–Goldfarb–Shanno Quasi-Newton, Levenberg-Marquardt algorithms. The also includes fuzzy layer identify best function based performance. Publicly available cleaned versions NASA datasets study. Various measures evaluation including specificity, precision, recall, F-measure, an area under receiver operating characteristic curve, accuracy, R2, mean-square error. Two graphical user interface tools developed MatLab implement framework. first tool comparing functions well extracting results; second selection using logic. A BR algorithm selected it outperformed others most measures. accuracy compared other widely techniques, from which was found better among all functions.