An Inclusive Hybrid Approach for Predicting Defects in Microservices Architecture Across Languages

Abstract

In the evolving landscape of software development, where monolithic frameworks are giving way to microservices-based architectures, a significant challenge lies in crafting a unified defect prediction model that transcends the boundaries of diverse programming languages, all within the context of continuous integration and continuous delivery (CI/CD) pipelines. This paper introduces a novel hybrid machine learning approach aimed at elevating the accuracy of defect prediction by seamlessly amalgamating disparate data sources and employing a diverse set of algorithms. The ultimate objective is the creation of a defect prediction model that is both language-agnostic and project-independent.

This hybrid model amalgamates Bidirectional Long Short-Term Memory (BiLSTM) networks with Attention mechanisms, static code metrics, and BERT-based language models. BiLSTM-Attention adeptly captures temporal dependencies residing within Abstract Syntax Trees (ASTs), while static code metrics furnish crucial insights into software complexity. Simultaneously, BERT lends its prowess in comprehending the textual context, thus facilitating a holistic comprehension of code snippets.

The research methodology encompasses a rigorous quantitative approach, commencing with an exhaustive literature review to establish a solid theoretical foundation. Subsequently, an empirical study unfolds, encompassing the entire gamut of activities ranging from data collection, preprocessing, and feature engineering, to model development, training, evaluation, analysis, validation, and the eventual derivation of conclusions. The insights derived from this research endeavour aspire to advance defect prediction techniques, thereby contributing significantly to the overarching goals of software engineering—namely, the pursuit of enhanced software quality and reliability.