Utilizing Machine Learning to Support Testing of Component Based Systems

Sammanfattning

This thesis investigates how machine learning (ML) can be integrated into the testing of component-based systems (CBSs), with a focus on enhancing integration testing through anomaly detection. A modular, agent-based testing framework was developed, combining Java-based simulation with a Python-implemented fault detection model to identify anomalies in system behavior without predefined failure thresholds. The framework leverages unsupervised learning, RESTful APIs, and multi-threaded execution to enable scalable and context-resilient testing over multiple rounds of testing. Through comparative analysis with traditional rule-based testing methods, the study demonstrates that ML-based approaches have the potential to offer superior adaptability and fault detection, particularly in scenarios with limited contextual information. The results contribute to ongoing efforts in software quality assurance by highlighting the role of intelligent, data-driven techniques in improving the efficiency and reliability of testing in modern, dynamic CBS environments.