Decoding the safety matrix: A conceptualisation of safety indicator-based variables for highway prediction models

Abstract

Despite the success of machine learning (ML) in safety risk prediction across various industries, there is scant evidence to prove that variables used are reliable indicators of safety performance. This study provides a conceptual framework for identifying safety indicators and formulating variables from these indicators that will optimise the quality of data used for risk modelling for Highways Traffic Officers (HTOs). This aligns with Sustainable Development Goal (SDG) 3: Good Health and Well-being, particularly target 3.6, which aims to halve global road traffic deaths and injuries. A mixed philosophical stance was adopted to understand and interpret the literature on safety indicators (SI) from myriad perspectives. A three-phase iterative waterfall approach was adopted which includes: i) PRISMA-based bibliometric search to identify relevant literature; ii) scientometric and cluster analysis to identify significant SIs and key considerations for selecting appropriate indicators for different purposes; iii) grounded theory analysis to synthesise scientific discoveries. Literature on leading and lagging SIs identified pertinent considerations that must be made when selecting SIs for risk modelling (e.g., action and utility). Furthermore, leading and lagging indicators were combined to form resilient indicators that incorporate adaptability into the safety system-thereby increasing safety performance. This paper presents a novel conceptual framework for HTOs which inculcates resilient measures in SI selection for adaptability and recovery purposes; and a robust method of integrating leading and lagging indicators to create resilience. Cumulatively, the research presents the first study to provide detailed guidance on the specific criteria of ML variables significant for risk modelling and prediction.