SOFTWARE DEFECT PREDICTION USING VERSION CONTROL DATA

• Main Author: FERDIAN SYAHPUTRA PANJAITAN (NIM: 23514030), ADI
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/20733
• Institution: Institut Teknologi Bandung

Software defect is a error, flaw, or fault in the software that causes the software fails to meet software requirements specifications or fails to meet user expe- ctations. Defect can not be avoided. Defect arise from mistakes and errors in either a source code or it’s design. It will take 50% - 70% from the total cost to do a defect fixing. Meanwhile, the software quality will be decreased. Defect can be prevented by utilizing the prediction model. This research focus is to propose defect prediction technique by using version control and source code data. Google Chromium will be used as data source as it licensed as open source software. That means source code and version control data can be accessed. Chromium also has a good documentation and standard on it’s development lifecycle. There’s 2 (two) kind of technique that will be used in this research. Defect prediction on file changes and prediction on author commit to see if there’s a correlation between software defect and author con- tribution. On the first experiment, the results shows that Naive Bayes give a better performances than the other algorithm. But the results overall is not good enough to give a prediction. The second prediction author can be ranked by it’s prediction probability to make a defect on software.