AN ADAPTIVE MODEL FOR SOFTWARE CODE QUALITY ASSESSMENT IN REFACTORING TASKS BASED ON FUZZY LOGIC
Annotatsiya
The article's objective is to develop a hybrid adaptive model for assessing software code quality based on code smell characteristics by combining fuzzy logic and machine learning methods to enhance the objectivity and efficiency of refactoring. The methodology underlying this research is aimed at developing a hybrid adaptive model for software code quality assessment. It combines fuzzy logic and artificial intelligence methods, specifically an adaptive neuro-fuzzy inference system (ANFIS). The multi-layered ANFIS implements the Takagi-Sugeno fuzzy inference with the ability to learn using gradient methods. The methodology is based on a hybrid approach that integrates expert knowledge with the automated training of the model on real data. Results. The research resulted in the development of a hybrid adaptive model for software code quality assessment based on fuzzy logic and the ANFIS. This model allows for automated, objective, and flexible code quality assessment in refactoring tasks. The model uses eight key code smell metrics: WMC, DIT, RFC, LCOM, NOA, NOC, CBO, and FANOUT. Their normalization and processing are performed using fuzzy logic based on the Takagi-Sugeno algorithm. This ensures that the uncertainty and subjectivity of expert evaluations are taken into account. The ANFIS architecture allows the model to learn from real data, with subsequent automated adjustment of the membership function parameters and rule weights. This enables the model to adapt to various technology stacks and projects. The use of trapezoidal membership functions increases the accuracy of modeling critical code smell zones, while the hybrid learning algorithm based on gradient descent ensures high precision in determining code quality, ultimately contributing to improved software efficiency, maintainability, scalability, and security. The scientific novelty of the research lies in the development of a hybrid adaptive model for software code quality assessment. Unlike existing models, this one is based on fuzzy logic and an ANFIS, which combines expert knowledge with automated training on real data to enhance the objectivity and efficiency of the refactoring process. The proposed ANFIS architecture with trapezoidal membership functions is used to process eight key code smell metrics (WMC, DIT, RFC, LCOM, NOA, NOC, CBO, FANOUT) within the context of Takagi-Sugeno fuzzy inference. This provides a flexible, interpretable, and adaptive assessment of code quality with the ability to automatically tune model parameters based on gradient learning, which significantly increases the accuracy of code quality determination and the model's suitability for various technology stacks and projects. The practical significance of the research lies in the direct implementability and integration of the developed hybrid adaptive model for software code quality assessment into existing static analysis tools and DevOps processes, specifically as plugins for Continuous Integration/Continuous Delivery (CI/CD) systems. This will enable automated, objective, and adaptive monitoring of code quality in real time. In addition, the model has significant potential for extension to various programming languages and technology stacks by analyzing large datasets from open-source repositories, which will enhance its universality and accuracy. A promising direction for future work is to improve the ANFIS architecture by incorporating deep learning methods, which would allow for the automatic detection of new code smells and their interdependencies. The development of interpretable mechanisms to explain the model's decisions will increase developer trust in the system and promote its widespread adoption in both industrial software development and educational processes in software engineering and cybersecurity.