Retrospective and predictive analysis of human operator performance with event report data of a nuclear reactor

Human Reliability Analysis (HRA) quantifies the likelihood of human operator's erroneous actions towards evaluation of risk emanating from complex systems, in terms of Human Error Probability. The main contribution of this article is in terms of the interactions between retrospective and predictive analysis of operator performance using insights from the recommendations of developing third-generation methods. This article highlights, (1) retrospective analysis lays the foundation of a good predictive analysis for HRA, and reciprocally, (2) a good predictive HRA method should be a replication of a robust retrospective analysis. In order to demonstrate this idea, we present a tool—APPROP (Application for Predictive and Retrospective analysis of Operator Performance). APPROP is a web tool and repository, based on features of Cognitive Reliability and Error Analysis Method and existing methods such as Standardized Plant Analysis Risk HRA (SPAR-H), to help practitioners in HRA data processing through debriefing. Using APPROP, a retrospective analysis was performed on event report data (2006–2020) from an operating nuclear reactor. The retrospective analysis scheme of APPROP enables HRA data processing from multiple sources and facilitates its classification into appropriate categories of context, error modes and error causes. A preliminary quantitative analysis with event report data gathered through APPROP was done using Logistic Regression (LR), Artificial Neural Networks (ANN) and Support Vector Machines based approaches. Preliminary predictive analysis results demonstrate that the LR and ANN-based models have the potential to perform predictive analysis and emulate the retrospective analysis.