The importance of replications in software engineering
Clauirton Siebra, Michael A. B. Mello · 2015
Prediction of defects in software is an important investigation area in software engineering, since such technique is able to return indications of parts of the code that are prone to contain problems. Thus, test teams can optimize the allocation of their resources by directing them to modules that are more defect-prone. The use of supervised learning is one of the approaches to support the design of prediction models. However, the erroneous use of training datasets can lead to poor models and, consequently, false results regarding accuracy. This work replicates important experiments of the area and shows how they could provide reliable results via the use of simple techniques of pre-processing. Based on the results, we discuss the importance of replications as method to find problems in current results and how this method is being motivated inside the software engineering area.