TY - GEN

T1 - Using different characteristics of machine learners to identify different defect families

AU - Petric, Jean

PY - 2016/6/1

Y1 - 2016/6/1

N2 - Background: Software defect prediction has been an active area of research for the last few decades. Many models have been developed with aim to find locations in code likely to contain defects. As of yet, these prediction models are of limited use and rarely used in the software industry.Problem: Current modelling techniques are too coarse grained and fail in finding some defects. Most of the prediction models do not look for targeted defect characteristics, but rather treat them as a black box and homogeneous. No study has investigated in greater detail how well certain defect characteristics work with different prediction modelling techniques.Methodology: This PhD will address three major tasks. First, the relation among software defects, prediction models and static code metrics will be analysed. Second, the possibility of a mapping function between prediction models and defect characteristics shall be investigated. Third, an optimised ensemble model that searches for targeted defects will be developed.Contribution: A few contributions will yield from this work. Characteristics of defects will be identified, allowing other researchers to build on this work to produce more efficient prediction models in future. New modelling techniques that better suit state-of-the-art knowledge in defect prediction shall be designed. Such prediction models should be transformed in a tool that can be used by our industrial collaborator in the real industry environment.

AB - Background: Software defect prediction has been an active area of research for the last few decades. Many models have been developed with aim to find locations in code likely to contain defects. As of yet, these prediction models are of limited use and rarely used in the software industry.Problem: Current modelling techniques are too coarse grained and fail in finding some defects. Most of the prediction models do not look for targeted defect characteristics, but rather treat them as a black box and homogeneous. No study has investigated in greater detail how well certain defect characteristics work with different prediction modelling techniques.Methodology: This PhD will address three major tasks. First, the relation among software defects, prediction models and static code metrics will be analysed. Second, the possibility of a mapping function between prediction models and defect characteristics shall be investigated. Third, an optimised ensemble model that searches for targeted defects will be developed.Contribution: A few contributions will yield from this work. Characteristics of defects will be identified, allowing other researchers to build on this work to produce more efficient prediction models in future. New modelling techniques that better suit state-of-the-art knowledge in defect prediction shall be designed. Such prediction models should be transformed in a tool that can be used by our industrial collaborator in the real industry environment.

U2 - 10.1145/2915970.2915979

DO - 10.1145/2915970.2915979

M3 - Conference contribution/Paper

SN - 9781450336918

BT - EASE '16 Proceedings of the 20th International Conference on Evaluation and Assessment in Software Engineering

PB - ACM

CY - New York

ER -