Predição de defeitos just-in-time em software utilizando inteligência artificial

Abstract

In the development or modification of a software, the product must have least amount of possible errors. Methods of predicting defects in software could be used for this. The principal objectives about this are performance a study and propose a defect prediction software Just-In-Time (JIT). Some advantages of the JIT approach are faster analise, better team utilization, easier identification of possible areas of code that are defective, and ease of finding the author (s) of modifications. In this work we present a proposal of the use of Just-In-Time (JIT) error identification using Artificial Neural Network (ANN) and decision tree (DT). The databases used as training, testing and validation have 227417 commits in total divided into six open source projects (Bugzilla, Columba, JDT, Mozilla, Platform and Postgres). The results obtained with techniques, ANN and DT, are on average higher than the works of comparation. The techniques used in the work development, as well as their similarities and differences with the previous approaches will be presented.