An Empirical Study on Predicting Software Development Bugs Using Dynamic Bayesian Networks

This study uses dynamic Bayesian networks (DBNs) to predict the number of bugs occurring during software development, based on software development metrics. We used data collected from multiple projects between 2016 and 2020, including person hours and the number of bugs during development, to construct DBNs for each project to predict the number of bugs occurring during development and then analyzed the accuracy of these predictions. This study evaluates a method to predict the number of bugs during development using information from the previous release and the target release. In addition, it compares the evaluation results obtained using DBNs with those obtained using traditional Bayesian networks. The dataset used was provided by e-Seikatsu, which supports the real estate industry with cloud services development projects.