A Deep Learning-based Approach for Predictive Evaluation of Microservice Maintainability

Abstract

Microservice Architecture (MSA) has emerged as a prominent paradigm in software system design, emphasizing decomposing monolithic applications into independent and modular functional services. This architectural approach provides a number of benefits; however, realizing these benefits requires a robust evaluation strategy focused on assessing the quality of the software system. This study proposes an innovative learning-based approach to evaluate the microservices’ quality, particularly maintainability. It is based on a deep learning technique that predicts the maintainability levels of microservices into three categories: low, medium, and high, with the low category indicating the need for refactoring. The prediction technique asses maintainability by feeding source-code metric values from different open-source microservice projects as inputs and obtaining results directly through transfer learning. The proposed method employed transfer learning and achieved % a 91.83 F1 score on the validated dataset obtained from open-source projects for predicting the need for refactoring services. Additionally, it reached %82 overall accuracy in the three class categorizations, showing notable performance. Considering these findings, it can be stated that the proposed learning-based evaluation is an effective method to assess microservice quality. As a result, this proposed method offers valuable insights for informed decision-making in software development and maintenance processes.