Applying Machine Learning to Risk Assessment in Software Projects

• Tipo: Artigo em Livro de Atas de Conferência Internacional
• Data: 2021

Autores

Sousa, A (Autor) Outra; João Pascoal Faria (Autor) FEUP; João Mendes-Moreira (Autor) FEUP; Gomes, D (Autor) Outra; Henriques, PC (Autor) Outra; Graca, R (Autor) Outra

Ata de Conferência Internacional

Título: MACHINE LEARNING AND PRINCIPLES AND PRACTICE OF KNOWLEDGE DISCOVERY IN DATABASES, PT II

Páginas: 104-118

21st Joint European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD)

ELECTR NETWORK, SEP 13-17, 2021

Indexação

ISI Web of Knowledge - 1 Citação

Scopus - 6 Citações

Outras Informações

ID Authenticus: P-00W-5PW

DOI: 10.1007/978-3-030-93733-1_7

Abstract (EN): Risk management is one of the ten knowledge areas discussed in the Project Management Body of Knowledge (PMBOK), which serves as a guide that should be followed to increase the chances of project success. The popularity of research regarding the application of risk management in software projects has been consistently growing in recent years, especially with the application of machine learning techniques to help identify risk levels of risk factors of a project before its development begins, with the goal of improving the likelihood of success of these projects. This paper presents the results of the application of machine learning techniques for risk assessment in software projects. A Python application was developed and, using Scikit-learn, two machine learning models, trained using software project risk data shared by a partner company of this project, were created to predict risk impact and likelihood levels on a scale of 1 to 3. Different algorithms were tested to compare the results obtained by high performance but non-interpretable algorithms (e.g., Support Vector Machine) and the ones obtained by interpretable algorithms (e.g., Random Forest), whose performance tends to be lower than their non-interpretable counterparts. The results showed that Support Vector Machine and Naive Bayes were the best performing algorithms. Support Vector Machine had an accuracy of 69% in predicting impact levels, and Naive Bayes had an accuracy of 63% in predicting likelihood levels, but the results presented in other evaluation metrics (e.g., AUC, Precision) show the potential of the approach presented in this use case.

Idioma: Inglês

Tipo (Avaliação Docente): Científica

Nº de páginas: 15