View in EDS

Evaluating Code Quality of AI-generated Mobile Applications : A Comparative Study of React Native and Kotlin Implementations ; Utvärdering av kodkvalitet i AI-genererade mobilapplikationer : En jämförande studie av React Native- och Kotlin-implementeringar

Saved in:
Bibliographic Details
Title: Evaluating Code Quality of AI-generated Mobile Applications : A Comparative Study of React Native and Kotlin Implementations ; Utvärdering av kodkvalitet i AI-genererade mobilapplikationer : En jämförande studie av React Native- och Kotlin-implementeringar
Authors: Wehbi, Nathalia, Jönsson, Axel
Publisher Information: Blekinge Tekniska Högskola, Institutionen för programvaruteknik
Publication Year: 2025
Collection: BTH (Blekinge Institute of Technology): DIVA / Blekinge Tekniska Högskola
Subject Terms: AI-generated code, AI-assisted development, Large Language Models in coding, Code quality, Native and Non-Native applications, Software Engineering, Programvaruteknik
Description: The increasing integration of AI-powered tools in software development raises crucial questions about the quality of the code they generate, particularly in rapidly evolving fields like mobile application development. This study addresses the need for up-to-date evaluations of AI-generated code quality in non-native applications, a gap in current research. To investigate this problem, we conducted an experiment where five prominent AI code generation tools– Gemini Code Assist, GitHub Copilot, ChatGPT, Windsurf IDE, and Deepseek– were prompted to generate code for a chess game in two mobile development frameworks: React Native and Kotlin. This resulted in a comparative analysis of ten AI-generated applications. The quality of the generated code was assessed using software quality metrics, informed by a comprehensive literature review. Our analysis revealed a moderate to high degree of variation across the generated applications in key metrics such as cyclomatic complexity, lines of code, and cognitive complexity. However, the observed results did not provide conclusive evidence to definitively identify a single AI tool as consistently producing the highest quality code across both frameworks. While the study provides valuable insights into the variability of code quality among different AI tools, the findings suggest that further research is necessary to achieve a more comprehensive understanding of the factors influencing the quality of AI-generated code. More in-depth investigation is required to draw definitive conclusions regarding the optimal AI tools for specific development contexts and to explore strategies for consistently generating high-quality code with AI assistance.
Document Type: bachelor thesis
File Description: application/pdf
Language: English
Availability: http://urn.kb.se/resolve?urn=urn:nbn:se:bth-28142
Rights: info:eu-repo/semantics/openAccess
Accession Number: edsbas.18D3C016
Database: BASE
Description
Abstract:The increasing integration of AI-powered tools in software development raises crucial questions about the quality of the code they generate, particularly in rapidly evolving fields like mobile application development. This study addresses the need for up-to-date evaluations of AI-generated code quality in non-native applications, a gap in current research. To investigate this problem, we conducted an experiment where five prominent AI code generation tools– Gemini Code Assist, GitHub Copilot, ChatGPT, Windsurf IDE, and Deepseek– were prompted to generate code for a chess game in two mobile development frameworks: React Native and Kotlin. This resulted in a comparative analysis of ten AI-generated applications. The quality of the generated code was assessed using software quality metrics, informed by a comprehensive literature review. Our analysis revealed a moderate to high degree of variation across the generated applications in key metrics such as cyclomatic complexity, lines of code, and cognitive complexity. However, the observed results did not provide conclusive evidence to definitively identify a single AI tool as consistently producing the highest quality code across both frameworks. While the study provides valuable insights into the variability of code quality among different AI tools, the findings suggest that further research is necessary to achieve a more comprehensive understanding of the factors influencing the quality of AI-generated code. More in-depth investigation is required to draw definitive conclusions regarding the optimal AI tools for specific development contexts and to explore strategies for consistently generating high-quality code with AI assistance.