Automated prioritization of construction project requirements using machine learning and fuzzy Failure Mode and Effects Analysis (FMEA)

Abstract

Due to the emergence of risk-based verification practices, requirement prioritization has gained importance as a task in project requirements management. However, it is challenging due to the voluminous paper-based requirements, the reliance on manual content analysis and interviews, the limited focus on specific requirement types, and the neglect of the non-conformance detectability factor. This paper leverages historical data of risk priority numbers (RPN) for many written requirements to develop a novel machine learning model for requirement prioritization. The training data includes requirement statements manually annotated with professionals' ratings for three risk factors severity, probability, and non-detectability. The overall fuzzy RPN data was aggregated using fuzzy logic to tackle uncertainty in subjective ratings. The model was trained using a Convolutional Neural Network (CNN). The study is anticipated to greatly aid professionals in prioritizing extensive textual project requirements essential for effective resource allocation that can eventually eliminate costly consequences of high-risk non-compliance.