| Software has become ubiquitous in our daily lives, and software with trusted quality has been critical in many domains. As a result, software quality assurance has been a heated topic for several decades. More precise quality management can be accomplished if we can allocate resources based on accurate quality estimation in the early stages of a project. Many software metrics and software quality estimation models have been proposed to enhance software quality and they have shown promise, but their usefulness was controversial and none of them was widely accepted. Moreover, there is no practical guideline for building software quality estimation models.;The proposed procedure is applicable to build defect estimation models from scratch or improve performance of current algorithmic estimation models. Moreover, it can be used in other areas of software estimation such as software cost estimation.;Keywords: software quality, quality estimation, statistical techniques, soft computing, neuro-fuzzy approach.;In this dissertation, a general procedure is proposed to derive software defect estimation models and various techniques are presented to accomplish the tasks in individual steps. The description of the six-step procedure is clearly stated and thorough. The first step is to review the literature review in order to identify the estimation target and related problems. Next, data preparation is done to collect suitable data for the modeling. Then statistical techniques together with machine learning methods are used to validate the effectiveness of software metrics. After that, statistical modeling can be performed to obtain the regression formula. Moreover, a neuro-fuzzy recalibration approach is adopted to improve the accuracy of the statistical model. Lastly, the evaluation of the derived model can be made. This six-step procedure is carried out based on data from both the ISBSG repository and OSS projects to present its practical value. The results demonstrate the significance of the procedure in building software quality estimation models. |
