Predicting Bug-Fixing Time with Machine Learning - A Collaborative Filtering Approach

Abstract

Predicting bug-fixing time helps software managers and teams prioritizetasks, allocations and costs in software projects. In literature,machine learning (ML) models have been proposed to predict bugfixingtime. One of features highlighted by studies is the reporter(the person who open the bug) has positive influence in the timeto resolve a bug. In this way, this paper answers the following researchquestion: How does a collaborative filtering approach performin predicting bug-fixing time compared to the supervised machinelearning approaches? In order to answer this question we performedan experiment using collaborative filtering approach to recommendthe bugs that are fast to be resolved in two open software projects.We compare our proposed approach with the ML approach relatedto the literature. As a result, the collaborative filtering approachoutperforms the supervised ML achieving an F-measure of 74%while the supervised ML achieved 66%. The collaborative filteringapproach showed to be a new perspective to predict bug-fixing timein software projects focusing the prediction on the reporter.

Predicting Bug-Fixing Time with Rating Features - A Comparison Between Ratings and Reputations

“…To answer the research questions, we carried out an experiment using the Eclipse Platform and NetBeans datasets. These datasets are available in MSR 2011 [14] and the work of Rodrigues and Parreiras [7]. Eclipse and NetBeans are two popular open-source projects for integrated development environments (IDE).…”

“…Google Colab is an environment prepared to execute data analyses directly in a browser 1 . We selected ML algorithms present in bug-fixing time prediction literature: Random Forest, Logistic Regression, KNN, Naive Bayes Gaussian, Naive Bayes Multinomial, SVM, MultiLayer Perceptron, SGD Classifier and Gradient Boosting [3,7,10,15].…”

“…In order to evaluate our features, an experiment was performed using bug tracking data from the following systems: Eclipse and Netbeans available from MSR 2011 [14] and used in the work of Rodrigues and Parreiras [7]. Only bug reports with the resolution status of "FIXED" were used in this experiment.…”

“…Studies have shown that the person who opens the bug has a positive impact on predicting bug-fixing time in ML models [3][4][5][6]. Rodrigues and Parreiras [7] have demonstrated that collaborative filtering can be utilized to predict bug-fixing time. In this approach, the users (reporter and developer) are the focal point of predictions.…”

“…To address these inquiries, we expand upon the work of Rodrigues and Parreiras [7] by conducting an experiment using bug reports from the NetBeans project and Eclipse Platform as datasets. We applied the reputation and rating features in ten supervised ML techniques.…”