Short answer

Integrate advanced AI techniques, particularly transformer models, into software quality assurance workflows to achieve higher accuracy in identifying code smells, while being mindful of computational costs and recall limitations.

Field: Innovation & Design
Source: Applied Sciences (2025)
Method: Comparative analysis and model development
Evidence: Strong effect

Transformer-based models, like RABERT, can significantly improve automated code smell detection accuracy by capturing complex semantic dependencies in software structures, outperforming traditional machine learning methods. This innovation & design research insight is drawn from a 2025 study published in Applied Sciences. Using Comparative analysis and model development, researchers explored how this design variable affects real-world outcomes. The key design takeaway: Integrate advanced AI techniques, particularly transformer models, into software quality assurance workflows to achieve higher accuracy in identifying code smells, while being mindful of computational costs and recall limitations.

Study

Innovation & DesignNew This WeekStrong effect

Transformer Models Achieve 90% Accuracy in Automated Code Smell Detection

Applied Sciences · 2025

Key Findings

01RABERT achieved 90.0% accuracy and 91.0% precision in code smell detection.
02Transformer-based models surpass classical machine learning methods in capturing intricate semantic dependencies.
03Challenges include low recall (53.0%) and significant computational overhead for transformer models.

Application

Design takeaway

How to apply

Explore and implement transformer-based AI models for code analysis tools to enhance the identification of software defects and improve overall code quality.

Project actions

01When evaluating AI models, consider multiple performance metrics beyond just accuracy.
02Document the trade-offs between model complexity, performance, and computational resources.

Method & Evidence

AimCan transformer-based models with relational embeddings enhance automated code smell detection compared to classical machine learning approaches?

MethodComparative analysis and model development

ProcedureA novel transformer-based model (RABERT) integrating relational embeddings was developed and compared against classical machine learning models (gradient boosting, SVM) for code smell detection. Performance was evaluated using metrics like accuracy, precision, and recall.

ContextSoftware engineering and quality assurance

Variables

IV["Model architecture (Transformer-based vs. Classical ML)"]

DV["Accuracy","Precision","Recall","Computational efficiency"]

CV["Dataset of code snippets","Definition of code smells","Evaluation metrics"]

Strengths & Limitations

Strengths

+Introduction of a novel transformer-based model (RABERT).
+Comprehensive comparative analysis with established ML techniques.

Limitations

The study's findings on recall and computational overhead suggest that practical implementation might require further research into model optimization or alternative approaches.

Reliability & validity

The study's validity is supported by a comparative analysis against established methods. Reliability could be further enhanced by testing across diverse codebases and programming languages.

Think critically

Given the low recall rate, how might a design team mitigate the risk of missed code smells when relying on this AI detection method?

Design Principles

"Leverage deep learning architectures to model complex relationships within data for improved analytical outcomes in design and engineering tasks."

This advancement in AI-driven code analysis offers a more robust and efficient way to identify potential issues in software development. By automating the detection of code smells, design teams can proactively address maintainability and extensibility challenges, leading to higher quality software products and reduced long-term development costs.

What This Means for Your Design

New computer programs that use a type of AI called transformers can find bad code patterns (code smells) much more accurately than older methods, but they sometimes miss some bad patterns and need a lot of computer power.

How to use in your project

1.Cite this study when discussing the application of AI in design analysis or automated quality assessment.

Add to My Project

Quick Cite

Paragraph starter

This research demonstrates the potential of transformer-based AI models, such as RABERT, to significantly enhance automated code smell detection, achieving high accuracy (90.0%) and precision (91.0%) by effectively modeling complex software structures. While these advanced models outperform traditional machine learning techniques, challenges related to recall (53.0%) and computational demands necessitate further investigation for broader practical application in design projects.

Source

Applied Sciences

Enhancing Software Quality with AI: A Transformer-Based Approach for Code Smell Detection

journal · 2025

View source

Questions About This Research

What does the research say about transformer models achieve 90% accuracy in automated code smell detection?: Integrate advanced AI techniques, particularly transformer models, into software quality assurance workflows to achieve higher accuracy in identifying code smells, while being mindful of computational costs and recall limitations. Evidence: Applied Sciences (2025).
Why does "Transformer Models Achieve 90% Accuracy in Automated Code Smell Detection" matter for design?: This advancement in AI-driven code analysis offers a more robust and efficient way to identify potential issues in software development. By automating the detection of code smells, design teams can proactively address maintainability and extensibility challenges, leading to higher quality software products and reduced long-term development costs.
How can designers apply this research?: Integrate advanced AI techniques, particularly transformer models, into software quality assurance workflows to achieve higher accuracy in identifying code smells, while being mindful of computational costs and recall limitations.
What were the main findings?: RABERT achieved 90.0% accuracy and 91.0% precision in code smell detection.. Transformer-based models surpass classical machine learning methods in capturing intricate semantic dependencies.. Challenges include low recall (53.0%) and significant computational overhead for transformer models.
What research method was used?: Comparative analysis and model development.
How strong is the evidence?: Evidence strength is rated Strong effect, based on a 2025 journal from Applied Sciences.
What should I do differently in my next project?: Explore and implement transformer-based AI models for code analysis tools to enhance the identification of software defects and improve overall code quality.
What are the limitations?: Low recall rates and high computational overhead for transformer models may limit immediate widespread adoption without further optimization.