Augmenting the software testing workflow with machine learning

Abstract

This work presents the ML Software Tester, a system for augmenting software testing processes with machine learning. It allows users to plug in a Git repository of the choice, specify a few features and methods specific to that project, and create a full machine learning pipeline. This pipeline will generate software test result predictions that the user can easily integrate with their existing testing processes. To do so, a novel test result collection system was built to collect the necessary data on which the prediction models could be trained. Test data was collected for Flask, a well-known Python open-source project. This data was then fed through SVDFeature, a matrix prediction model, to generate new test result predictions. Several methods for the test result prediction procedure were evaluated to demonstrate various methods of using the system.