Test Data Selection Based on Applying Mutation Testing to Decision Tree Models

Abstract

Software testing is crucial to ensure software quality, verifying that it behaves as expected. This activity plays a crucial role in identifying defects from the early stages of the development process. Software testing is especially essential in complex or critical systems, such as those using Machine Learning (ML) techniques, since the models can present uncertainties and errors that affect their reliability. This work investigates the use of mutation testing to support the validation of ML applications. Our approach involves applying mutation analysis to the decision tree structure. The resulting mutated trees are a reference for selecting a test dataset that can effectively identify incorrect classifications in machine learning models. Preliminary results suggest that the proposed approach can successfully improve the test data selection for ML applications.