Insights into the Applications of Bayesian Networks in Software Engineering

[Context] Bayesian networks (BNs) have been employed as a promising technique for combining existing project data with expert knowledge to support decision-making and help solve various software engineering (SE) problems. Existing secondary studies have addressed BNs in specific contexts. However, no tertiary study has been conducted to identify and catalog individual secondary studies, synthesizing the existing evidence on applying BNs in SE (BN4SE) from a broader perspective. [Objective] This paper aims to synthesize and analyze the secondary studies published on BN4SE to provide insights and opportunities for researchers and practitioners. [Method] We conducted a tertiary study following the guidelines available in the SE literature. [Results] We identified seven secondary studies addressing BNs in SE: one dedicated to software effort prediction, two to software quality prediction, one to software project management, one to requirements engineering, one to software testing, and one to several problems as it targeted the broader SE field. New research opportunities may arise from challenging aspects of BN application, such as the definition of probabilities, use of expert knowledge, use of models in practice, tool support, data availability issues, and rationale for model construction. [Conclusion] Despite the significant number of available primary studies and the relevance of BNs for the SE area, the number of secondary studies on the topic is very limited. This sheds light on a range of opportunities for the maturation of the field, e.g., the application of BNs in underexplored areas, methodological approaches for exploiting expert knowledge in data-constrained scenarios, and guidelines for detailed model construction and reporting.