Machine learning nested MCDM model to enhance decision reliability for transport safety engineering

Abstract

Ensuring robust and defensible decision is a critical attribute of multi-criteria decision-making (MCDM) activities, particularly in public sector decision-making (e.g., transport safety engineering). To this end, this study introduces an advanced machine learning embedded MCDM model that integrates the preference selection index (PSI), an alternative ranking order method accounting for two-step normalization (AROMAN), and a Gaussian mixture model (GMM), i.e., PSI–AROMAN–GMM, aiming to provide a reliable decision support system in transport safety engineering. In particular, the proposed model incorporates a machine learning algorithm (i.e., t-distributed stochastic neighbor embedding (t-SNE)) to reduce the computational load and enable efficient handling of large datasets. Specifically, this approach addresses the challenge faced by conventional GMM in uncertain initialization and identification of distinct natural clusters, and it also resolves the issue of GMM in setting correct number of Gaussian components to avoid the model overfitting or underfitting. Through a case study on transport safety engineering for G20 countries, multilevel empirical comparisons validate the robustness of the proposed model, highlighting its practicality and efficiency in informing reliable decisions and policy insights. Overall, this study provides decision-makers, practitioners, and engineers with a comprehensive framework for handling real-world socio-economic activities, especially in transport safety engineering across varied national contexts, with substantial reliability and applicability.