Decision support for global software development with pattern discovery
Abstract
Background: Software development process nowadays is becoming more globalized than ever before. Global Software Development (GSD) implies that the software development process is spread across countries and geographic boundaries. GSD brings challenges to software project leaders I managers because of the increase in management difficulty. As a result, utilizing data mining and machine learning techniques to provide quantitative, objective and predictive solution for project management is essential. Aim: To facilitate software project However, most of the previously employed techniques act as a black-box which represents results that are difficult for domain experts to examine. Two main problems are: (i) they require parameters calibration for accurate prediction and (ii) they focus on estimation accuracy quantitatively while qualitative information like "What are the most influential elements in the data that lead to the particular quantitative prediction?" are difficult to obtain by the practitioners (i.e., project leaders). management to make decisions by mining embedded knowledge from data and providing meaningful results. Method: In this paper we propose to adopt a pattern discovery technique which has been successfully applied in the field of computational Biology. The technique discovers association patterns inherited in the data which can provide insightful information for domain experts (e.g., project leaders), therefore increasing their confidence in making decisions. We apply the technique in the software defect datasets from the NASA MDP repository to predict whether a software project is defective or not and find out important factors in the data that signaled the prediction. In this paper we propose to adopt a pattern discovery technique which has been successfully applied in the field of computational Biology [7]. The technique was proposed to classify cancerous tissues by mining association patterns in the microarray data. We adopt the technique in this study for defective software projects detection in the datasets from the NASA MDP repository [8]. Association rule mining is the discovery of association relationships amo