Data-driven human error accident risk control: A systematic review of data sparsity, causal explainability, and operational transformation

With the increasing complexity of modern technological systems, data-driven human error accident risk control (DD-HEARC) faces three interrelated core challenges: data sparsity (Q1), explainability (Q2), and transformation application (Q3), which form a vicious cycle that hinders scientific progress. Following the PRISMA guidelines, this study systematically reviewed 95 high-quality journal articles from Web of Science, Scopus, and IEEE Xplore. The analysis reveals severe fragmentation: approximately 15% of studies address two or more challenges concurrently, and only 21% explicitly incorporate causal inference. Four key limitations are identified: causal blindness in data augmentation, disconnection between explainable AI and safety practice, lack of systematic transformation mechanisms, and absence of cross-dimensional integration methodologies. The study's core contribution is the systematic diagnosis of the Q1→Q2→Q3→Q1 constraint loop and the proposal of an innovative three-dimensional DD-HEARC framework integrating temporal, logical, and collaborative dimensions. This framework provides a unified cognitive tool for researchers and practitioners to shift from fragmented, experience-driven approaches toward systematic, science-driven risk governance.