Supervised Learning for Test Suit Selection in Continuous Integration

Continuous Integration is the process of merging code changes into a software project. Keeping the master branch always updated and unfailingly is very computationally expensive due to the number of tests and code that needs to be executed. The waiting times also increase the time required for debugging. This paper proposes a solution to reduce the execution time of the testing phase, by selecting only a subset of all the tests, given some code changes. This is accomplished by training a Machine Learning (ML) Classifier with features such as code/test files history fails, extension code files that tend to generate more errors during the testing phase, and others. The results obtained by the best ML classifier showed results comparable with the recent literature done in the same area. This model managed to reduce the median test execution time by nearly 10 minutes while maintaining 97% of recall. Additionally, the impact of innocent commits and flaky tests was taken into account and studied to understand a particular industrial context.