Predicting number of bugs before launch: An investigation based on machine learning

Document Type

Conference Proceeding

Publication Date

7-1-2021

Abstract

Identifying and minimizing the number of bugs before release is a high priority of any team working on software development. This can be achieved by Machine Learning (ML) models By using particular aspects of the code, ML Models can predict the number of bugs that are possible post launch. We use a public dataset consisting of 15 Java projects from GitHub as our training and test dataset. We use five ML models for our investigation: Multilayer Perceptron, K-Nearest Neighbors, Linear Regression, Logistic Regression, and Decision Trees We conduct a preliminary investigation to evaluate how these ML models perform in predicting bugs. The results show that Linear Regression outperforms the other four ML models in finding the number of bugs post release.

Publication Title

Proceedings - 2021 IEEE 45th Annual Computers, Software, and Applications Conference, COMPSAC 2021

First Page

1403

Last Page

1404

Digital Object Identifier (DOI)

10.1109/COMPSAC51774.2021.00205

ISBN

9781665424639

Share

COinS