Safety and Security Analysis using LDA based on Case Reports: Case Study and Trust Evaluation Method

Abstract

There are many cases where the safety and security of systems are threatened by accidental or intentional human error. This study focuses on the fact that there is information available about human error in design and operation documents and case reports, and they are in natural language. Therefore, we propose a method to analyze the impact of human error on safety and security using Latent Dirichlet Allocation (LDA), which is one of the topic model methods. First, we matched the given information to create a list of similarities (co-occurrence list) between documents. Based on this co-occurrence list, a fault and attack tree was constructed. While manually considering them, the critical points were identified through sensitivity analysis. We show the effectiveness of this proposed method through two characteristic case studies of cyber-based connected car design deficiencies and physical-based manufacturing inspection fraud. Both analyzes add a way to leverage big data interoperability in manufacturing processes using the IoT.