Machine learning-based approach for code smell detection in software JavaScrip

• Ano de defesa: 2024
• Autor(a) principal: Sarafim, Diego Santana
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Aprendizado de Máquina; Classificação; Classification; Code smells; Code Smells; Conjunto de Dados; Dataset; JavaScript; Machine learning
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/100/100131/tde-13122024-002027/

JavaScript has become one of the world\'s most widely used programming languages. It is characterized by being a dynamic, interpreted, and weakly typed scripting language, which is especially suitable for the development of web applications. While these features allow the language to offer high levels of flexibility, they can also make JavaScript code more challenging to write, maintain, and evolve. One of the risks that JavaScript and other programming languages are prone to is the presence of code smells. Code smells result from poor programming choices during source code development, which negatively influence the source code comprehension and maintainability in the long term. %Research on the topics related to code smells has helped elevate our comprehension of the factors that lead to code smell incidence and has led us to a better understanding of their nature and their risks. Part of the research on code smells was focused on exploring different techniques to detect them. The detection techniques explored over the years range from using manually designed heuristics to employing machine learning algorithms. Although presenting promising results, most of the studies that utilized machine learning techniques to detect code smells focused on Java programming language. Since different programming languages can have different syntax, purposes, and structures and can support different programming paradigms, they may pose varying challenges for smell detection and can even be prone to unique code smells. JavaScript is such a programming language that differs from Java in a few fundamental ways, among these differences, the syntax, the support for different programming paradigms and the vulnerability to unique types of code smells stand out. Encouraged by an apparent lack of studies exploring the employment of machine learning techniques to detect code smells in JavaScript, we conducted this work for which the main contributions are the construction of the first datasets for code smell detection on JavaScript and the application of machine learning techniques for the detection of code smells in JavaScript.