Performance indicators analysis in software processes using semi-supervised learning with information visualization

Abstract

Software development process requires judicious quality control, using performance indicators to support decision-making in the different processes chains. This paper recommends the use of machine learning with the semi supervised algorithms to analyze these indicators. In this context, this paper proposes the use of visualization techniques of multidimensional information to support the labeling process of samples, increasing the reliability of the labeled indicators (group or individual). The experiments show analysis from real indicators data of a software development company and use the algorithm bioinspired Particle Competition and Cooperation. The information visualization techniques used are: Least Square Projection, Classical Multidimensional Scaling and Parallel Coordinates. Those techniques help to correct the labeling process performed by specialists (labelers), enabling the identification of mistakes in order to improve the data accuracy for application of the semi-supervised algorithm.