Statistical Analysis for Revealing Defects in Software Projects

Dissertation presented as the partial requirement for obtaining a Master's degree in Information Management, specialization in Information Systems and Technologies ManagementDefect detection in software is the procedure to identify parts of software that may comprise defects. Software companies always seek to improve the performance of software projects in terms of quality and efficiency. They also seek to deliver the soft-ware projects without any defects to the communities and just in time. The early revelation of defects in software projects is also tried to avoid failure of those projects, save costs, team effort, and time. Therefore, these companies need to build an intelligent model capable of detecting software defects accurately and efficiently. This study seeks to achieve two main objectives. The first goal is to build a statistical model to identify the critical defect factors that influence software projects. The second objective is to build a statistical model to reveal defects early in software pro-jects as reasonable accurately. A bibliometric map (VOSviewer) was used to find the relationships between the common terms in those domains. The results of this study are divided into three parts: In the first part The term "software engineering" is connected to "cluster," "regression," and "neural network." Moreover, the terms "random forest" and "feature selection" are connected to "neural network," "recall," and "software engineering," "cluster," "regression," and "fault prediction model" and "software defect prediction" and "defect density." In the second part We have checked and analyzed 29 manuscripts in detail, summarized their major contributions, and identified a few research gaps. In the third part Finally, software companies try to find the critical factors that affect the detection of software defects and find any of the intelligent or statistical methods that help to build a model capable of detecting those defects with high accuracy. Two statistical models (Multiple linear regression (MLR) and logistic regression (LR)) were used to find the critical factors and through them to detect software defects accurately. MLR is executed by using two methods which are critical defect factors (CDF) and premier list of software defect factors (PLSDF). The accuracy of MLR-CDF and MLR-PLSDF is 82.3 and 79.9 respectively. The standard error of MLR-CDF and MLR-PLSDF is 26% and 28% respectively. In addition, LR is executed by using two methods which are CDF and PLSDF. The accuracy of LR-CDF and LR-PLSDF is 86.4 and 83.8 respectively. The standard error of LR-CDF and LR-PLSDF is 22% and 25% respectively. Therefore, LRCDF outperforms on all the proposed models and state-of-the-art methods in terms of accuracy and standard error