Predicting Post-Release Defects in OO Software using Product Metrics Gabriel de Souza Pereira Moreira 1 , Roberto Pepato Mellado 1 , Robson Luis Monteiro Junior 1 , Adilson Marques da Cunha 2 , and Luiz Alberto Vieira Dias 2 1 IMAGEM - Geographic Information Solutions, Software Engineering Department Sao Jose dos Campos, SP, Brazil {gspmoreira,rpepato,kimmaydesign}@gmail.com 2 ITA – Aeronautics Institute of Technology, Computer Science Division, Sao Jose dos Campos, SP, Brazil {cunha,vdias}@ita.br Abstract — Software maintenance has consumed more than 50% of develop- ment effort and about 90% of software lifecycle. Finding and correcting defects after software delivery have often presented high costs when compared to cor- rect it on previous project phases. Within this context, defect prediction has at- tracted growing interest from industry and academy. In this study, a survey is conducted, considering two object-oriented systems developed by industry and currently under maintenance. It was proposed and implemented a method for collection and integration of software product metrics for defect prediction. Code and design metrics, at class level, were extracted using static code analy- sis. The code modules where defects were detected were obtained from correc- tive maintenance history data. The prediction models presented are based on Multivariate Linear Regression and have used internal quality metrics as pre- dictors and detected defects as predicted variables. This approach can be used to help prioritizing quality activities like testing, inspecting, and refactoring on defect-prone classes. Keywords. Software corrective maintenance, defect prediction, defect prone- ness, defect volume, object-oriented, software metrics, multivariate regression analysis, iterative process. 1 Introduction As software lifetime increases, code and design qualities become important factors for development and maintenance of cost reduction. Kemerer and Slaughter attested that software maintenance is an understudied phe- nomenon within the research community [1]. They estimate that software mainte- nance activities can constitute 50% of all efforts undertaken in software development. Bennett reported in [2] that 40% to 90% of the software product lifetime total cost is spent in maintenance. Jones [3] stated that during the early years of the XXI centu-