Аннотация
This article examines the study of the modern market of algorithms and software products for working with Bayesian networks. One of the most important problems is the prediction of software flaws, which seems to be a necessary area in software development, because it helps creators to detect and eliminate difficulties before they turn into costly and difficult to implement errors. Early detection of software flaws focuses on saving time and money in the software development process and guarantees the nature of the final product. The purpose of this study is to analyze three algorithms of Bayesian network theory to classify whether a project is subject to defects. The selection is based on the fact that the most commonly used layout in the literature is naive Bayesian, but no Bayesian networks are used in any work. Thus, K2, Hill Climbing and TAN are used to build Bayesian networks. The results of various performance indicators used for cross-validation show that the results of systematization are comparable to a tree of conclusions and a disorderly forest, with the advantage that Bayesian algorithms show the least variability, which orients the technical software to have tremendous reliability in its forecasts, because the selection of training and testing information does not give unstable results