Integrated Metamorphic and Mutation Testing Approach for Deep Learning System Test Data Generation

Read full text Summarise with VIRAT
Authors: Radziah Mohamad, Junhan Li, Tan Ser Xuen, Johanna Ahmad; and Alif Ridzuan Khairuddin
Cited by: 0
References: 21

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This chapter presents an integrated testing framework that utilizes a metamorphic-and-mutation-based approach to produce high-quality test data for deep learning systems, as these systems are now being used in high-stakes domains such as healthcare, finance, and autonomous technologies, which consequently makes it important to maintain the reliability and integrity of their predictions. Conventional testing methods often struggle to address the challenge of complexity and non-determinism in deep learning models. Metamorphic testing employs pre-defined metamorphic relations to derive expected behavior from transformed inputs, enabling validation without explicit ground truth. Mutation-based testing, on the other hand, injects deliberate faults into the model or dataset to assess system robustness and uncover latent defects. This integrated approach enhances test coverage and supports the generation of diverse, adversarial, and edge-case data. By leveraging metamorphic relations and systematically injecting mutations into models and data, this approach addresses the oracle problem and uncovers hidden inconsistencies. The chapter highlights how these methods contribute to robust validation processes, ensuring integrity, reliability, and transparency in AI-driven applications. This aligns with the broader scope of promoting best practices for data quality, integrity, and assurance in intelligent systems.

Fields of Study & Subjects

Software Testing and Debugging Techniques Adversarial Robustness in Machine Learning Software System Performance and Reliability

Publication details

Journal / source
IntechOpen eBooks
Publication date
12 March 2026
Language
EN
Document type
Book chapter