Software Fault Proneness Prediction with Group Lasso Regression: On Factors that Affect Classification Performance

The Untold Impact of Learning Approaches on Software Fault-Proneness Predictions

“…Many different machine learning algorithms have been used in building software fault-proneness prediction models. These include J48 (Moser et al, 2008;Kamei et al, 2010;Krishnan et al, 2013), Random Forest (RF) (Guo et al, 2004;Mahmood et al, 2018;Fiore et al, 2021;Gong et al, 2021), and combinations of several machine learning algorithms, e.g., OneR, J48, and Naïve Bayes (NB) in (Menzies et al, 2007), RF, NB, RPart, and SVM in (Bowes et al, 2018), J48, RF, NB, Logistic Regression (LR), PART, and G-Lasso in (Goseva-Popstojanova et al, 2019), and Decision Tree (DT), k-Nearest Neighbor (kNN), LR, NB, and RF in (Kabir et al, 2021). With recent advances in Deep Neural Networks (DNN), some software fault-proneness prediction studies used deep learning (Wang et al, 2016;Li et al, 2017;Pang et al, 2017;Zhou et al, 2019;Zhao et al, 2021).…”

“…Static code metrics are collected from the software source code or binary code units (Koru and Liu, 2005;Menzies et al, 2007;Lessmann et al, 2008;Menzies et al, 2010;He et al, 2013;Ghotra et al, 2015;Bowes et al, 2018;Kabir et al, 2021). Change metrics, sometimes called process metrics, are collected from the projects' development history (i.e., commit logs) and bug tracking systems (Nagappan et al, 2010;Giger et al, 2011;Krishnan et al, 2011Krishnan et al, , 2013Goseva-Popstojanova et al, 2019). Social metrics are extracted from the communications among developers and/or users of a software project (Bird et al, 2009).…”

“…Social metrics are extracted from the communications among developers and/or users of a software project (Bird et al, 2009). Some studies used only static code metrics (Menzies et al, 2007;Mende and Koschke, 2009;Song et al, 2011;Xu et al, 2018Xu et al, , 2019Amasaki, 2020;Kabir et al, 2021), only change metrics (Giger et al, 2011;Krishnan et al, 2011Krishnan et al, , 2013Goseva-Popstojanova et al, 2019), organizational metrics (Nagappan et al, 2008) or metrics derived from the contribution networks (Pinzger et al, 2008). There are studies that used combinations of different types of metrics (Nagappan et al, 2010;Arisholm et al, 2010;Giger et al, 2012;Alshehri et al, 2018), including studies that combined metrics extracted from contribution networks, dependency networks, and/or socio-technical networks (Bird et al, 2009;Gong et al, 2021).…”

“…Works that do prediction at the file level use features collected from units such as files (Zimmermann et al, 2007;Moser et al, 2008;Kamei et al, 2010;Krishnan et al, 2013;Goseva-Popstojanova et al, 2019), classes (Koru and Liu, 2005;Malhotra and Raje, 2015;Song et al, 2019), or methods (Giger et al, 2012;Bowes et al, 2018), while prediction at the component level is based on features aggregated at the package, module, or component level from features extracted at file, class, or method level (Zimmermann et al, 2007).…”

“…Many papers have explored different factors that affect the performance of the classification-based software faultproneness prediction. These factors include: choice of machine learning algorithms (i.e., learners) (Lessmann et al, 2008;Krishnan et al, 2013), choice of software metrics (i.e., features used for prediction) (Bluemke and Stepień, 2016;Wang et al, 2016;Alshehri et al, 2018;Gong et al, 2021), effect of data balancing techniques (Khoshgoftaar et al, 2010;Wang and Yao, 2013;Song et al, 2019;Goseva-Popstojanova et al, 2019), and datasets used in building prediction models (Goseva-Popstojanova et al, 2019).…”