Reformative ROCOSD–ORESTE–LDA model with an MLP neural network to enhance decision reliability

• Published in: Knowledge-based systems 2024-02, Vol.286, p.111384, Article 111384
• Main Authors: Wang, Xiaoyuan, Hou, Bodong, Teng, Yuanhong, Yang, Yicheng, Zhang, Xinyue, Sun, Lei, Chen, Faan
• Format: Article
• Language: English
• Subjects: Decision reliability; Linear discriminant analysis; Multilayer perceptron; ORESTE; ROCOSD
• ISSN: 0950-7051; 1872-7409
• DOI: 10.1016/j.knosys.2024.111384

Multi-criteria decision-making (MCDM) problems require a decision model and outcomes that are stable and reliable, which is especially true for safety systems. To this end, we develop a hybrid MCDM model that combines robustness, correlation, and standard deviation (ROCOSD), organization, rangement et synthèse dedonnées relarionnelles (ORESTE), and linear discriminant analysis (LDA), namely, the ROCOSD–ORESTE–LDA model. In particular, we enhance the model performance by embedding a multilayer perceptron (MLP) neural network into the LDA to minimize the variance in the outcomes. Specifically, we address the possible model failure of a general LDA under non-ideal conditions such as shared mean of the distributions and non-Gaussian distributed samples. Based on a case study analyzing transport safety situations for G20 member countries over the past decade, the proposed model is shown to be adaptable, stable, and reliable via multiple experiments and multilevel comparisons. This systematic decision framework may aid in future transport safety development planning in G20 countries, and this methodology may be feasibly applied to resolve safety management issues, as well as other MCDM activities with high complexity and uncertainty.