Speaker: Khaled Al-Sabbagh
Abstract: Machine learning models have been increasingly used to support decision making in software engineering tasks. One example of its application is the optimization of test case selection in continuous integration. Among the challenges that hinders the application of machine learning is the amount of noise that comes in the data, which often leads to a decrease in classification performances. For this reason, we examine the impact of one type of noise, called class noise, on a learner’s ability for selecting test cases. Understanding the impact of class noise on the performance of a learner for test case selection would assist testers decide on the appropriateness of different noise handling strategies. The results show a statistically significant relationship between class noise and the learner’s performance for test case selection. We conclude that higher class noise in training data leads to missing out more tests in the predicted subset of test suite and increases the rate of false alarms when the level of class noise exceeds 30%.
Link to the recorded presentation on YouTube: