MACHINE LEARNING APPROACHES FOR AUTOMATIC TEST CASE GENERATION FROM REQUIREMENTS

The exponential growth in software complexity has intensified the demand for automated testing methodologies that can efficiently generate comprehensive test cases from natural language requirements. Traditional manual test case generation is labor-intensive, error-prone, and often fails to achieve adequate coverage of complex system behaviors. This paper presents a novel machine learning framework that leverages transformer-based language models and reinforcement learning techniques to automatically generate high-quality test cases directly from software requirements specifications. Our approach combines natural language processing (NLP) with semantic understanding to extract testable scenarios, boundary conditions, and edge cases from unstructured requirement documents. We introduce a hybrid architecture that integrates BERT-based requirement analysis with GPT-based test case synthesis, enhanced by a reinforcement learning component that optimizes test case quality through feedback mechanisms. Experimental evaluation on five industrial software projects demonstrates that our approach achieves 87.3% requirement coverage, 92.1% defect detection rate, and reduces manual test case creation time by 73%. The generated test cases exhibit superior fault detection capabilities compared to manually created test suites, with a 34% improvement in mutation score. Our contributions include: (1) a comprehensive taxonomy of requirement-to-test mappings, (2) a novel ML architecture for automated test generation, (3) extensive empirical validation across diverse domains, and (4) open-source tools for practitioners. The results indicate significant potential for transforming software testing practices through intelligent automation.

• Publication year

  2025

• Venue

  International journal of engineering science and advanced technology

• Publication date

  2025-11-01

• Fields of study

  Not labeled

• Identifiers
  DOI 10.64771/ijesat.2025.v25.i11.pp97-102
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• DART: directed automated random testing

  2005cited by this paper
• CUTE: a concolic unit testing engine for C

  2005cited by this paper

• No citing papers are available for this paper.