Prioritization of Regression Test Cases Based on Machine Learning Methods

Due to resource and time constraints involved in the software testing process, it is not possible to implement all test scenarios for each release. Test scenarios can be prioritized according to certain criteria defined by the developers to ensure effective execution of the testing process and detection of errors. This study investigated whether machine learning based models could be used to prioritize test scenarios created in regression testing. It is attempted to determine which tests can be prioritized for execution based on different independent variables. In total, each of the 964 test scenarios in the dataset was labelled as minor (482) and major (482) by two experts. In the models, the number of related requirements, the number of related errors, and the age of the scenario were used as independent variables, and the scenario classes labelled as minor - major were taken as the target variable. The scenarios were pre-processed using natural language processing techniques and different machine learning algorithms were used for model development. In the classification based on test scenarios, the random forest algorithm showed the best performance with a F1-score of 81%. In the classification based on the number of related requirements, the number of interrelated errors, and the age of the test scenarios, the random forest model once again demonstrated the highest success rate at 79%. This study demonstrates that machine learning techniques offer a variety of models for test case prioritization.