Bug Prediction Models: seeking the most efficient

Abstract

Choosing the most appropriate machine learning model for bug prediction tasks is critical. This paper primarily compares the predictive power of individual models versus ensemble models. We begin by experimenting with popular single-machine learning models commonly used in bug prediction, like neural networks and support vector machines. Additionally, we test with ensemble models that combine individual models' unique strengths, aiming to maximize each's benefits. Our evaluation is based on datasets containing historical development data from well-known open-source software projects. We rely on various metrics when assessing the models, encompassing accuracy, precision, recall, and F1 score. Based on our research findings, it has been observed that ensemble models tend to outperform single models, particularly when it comes to maintaining resilience across various datasets.Nevertheless, factors like the project's complexity, data availability, and computational resources all play a role in determining whether to use single or ensemble models. This paper offers a thorough analysis of the factors to consider when selecting machine learning models and approaches for bug prediction, providing valuable insights into the field. Furthermore, it offers practical advice for professionals, enabling them to make informed choices.