Exploring Metamorphic Testing for Self-learning Functions with User Interactions

This paper presents a novel approach to testing SLFs (SLFs) in adaptive systems using metamorphic testing (MT). Recognizing the challenge of verifying software that learns from and adapts to user behavior without a definitive oracle, we propose specific metamorphic relations (MRs) as the basis for our testing framework. These MRs are crafted to evaluate the SLFs’ capacity to tailor user experiences by adapting to individual behaviors, transitioning user patterns and environmental changes, and managing data anomalies. We demonstrate our testing methodology through a two-stage process, utilizing synthetic data generated by the CAGEN∗ tool to simulate realistic user interactions and environmental factors. Principal component analysis (PCA) is employed to visualize the effectiveness of SLFs in adhering to the identified MRs. Our findings highlight proficiency in personalization by differentiating between user behaviors. The paper emphasizes the effectiveness of MT in enhancing the development of intelligent, user-centric systems and suggests directions for future research to extend these testing methods to more complex scenarios and diverse SLF architectures