Test Case Creation Approach using LLM for Android System

• Bruno Cerdeira INDT
• Hermann Hernani INDT
• Luan Marques INDT
• Luiz Souza INDT
• Silvia Ascate INDT
• Eliane Collins INDT
• André Carvalho UFAM

DOI: https://doi.org/10.5753/cibse.2025.35328

Abstract

This research project presents a method to generate functional test cases for new Android features using the Large Language Model (LLM) LLama. The objective is to reduce manual effort and avoid process bottlenecks, benefiting software testing teams and developers. The approach uses LLM-generated prompts based on Android feature specifications for automated test case generation. A feasibility study evaluated 57 test cases, revealing that 19.3% were well-written, 63.2% had acceptable quality, and 26.3% had medium complexity. The research was conducted under contract with the Institute of Technological Development (INDT) and its funded by the R&D program of EMBRAPII for Motorola LTDA/Flextronic Ltda.

Keywords: Test Automation, Automatic Test Generation, Reinforcement Learning, AI, Android

References

Julia Jaremko, Dagmar Gromann, and Michael Wiegand. Revisiting implicitly abusive language detection: Evaluating llms in zero-shot and few-shot settings. In Proceedings of the 31st International Conference on Computational Linguistics, pages 3879–3898, 2025.

Roberto F Lima Jr, Luiz Fernando PB Presta, Lucca S Borborema, Vanderson N Silva, Marcio LM Dahia, and Anderson Santos. A case study on test case construction with large language models: Unveiling practical insights and challenges. In Congresso Ibero-Americano em Engenharia de Software (CIbSE), pages 388–395. SBC, 2024.

Zhe Liu, Chunyang Chen, Junjie Wang, Mengzhuo Chen, Boyu Wu, Xing Che, Dandan Wang, and Qing Wang. Chatting with gpt-3 for zero-shot human-like mobile automated gui testing. arXiv preprint arXiv:2305.09434, 2023.

Huynh Khanh Vi Tran, Michael Unterkalmsteiner, Jürgen Börstler, and Nauman bin Ali. Assessing test artifact quality - a tertiary study. Information and Software Technology, 139:106620, 2021.