The Study Of Software Defect-Fixing Effort Prediction Based On Empirical Data

Software defect-fixing effort prediction seeks to predict the effort needed to fix each defect prior to actual fixing activities. It's important to software industry because of its huge meanings for resource allocation, software process uncertainty reducing, and software quality assurance etc. The software defect-fixing effort prediction problem has not been well solved for its particularity and complexity.There are many disadvantages in existing defect-fixing effort prediction models, including high applying cost, low prediction precision, unable to conduct predictions in early stage of the defect lifecycle, and unable to conduct cross-project predictions. More investigations are needed for this problem.TCEPM is a defect-fixing effort prediction model based on text categorization techniques. It assumes that similar defects owning similar description text and will cost comparable fixing effort. By classifying the description text of defects using the SVM algorithm, TCEPM may predict the fixing effort needed for each detected defects. TCEPM introduces a three-step approach to perform predictions: 1) Preprocessing description text of defects in empirical data by using techniques such as Chinese word segmentation and text feature extraction; 2) Learning the prediction model form processed empirical data using SVM algorithm; 3) Predicting fixing effort for new submitted defects.Experiment results show that TCEPM is a good solution for software defect-fixing effort prediction with high practicability. Advantages of TCEPM include low applying cost, satisfied prediction results, predicting in the early stage of defect lifecycle, and applicable for cross-project prediction.