A Novel Relational Vector Method Integrated with Root Assessment Method for Evaluating Advanced Trauma Detection Models

This research proposes an innovative Multi-Criteria Decision Making (MCDM) framework for assessing trauma injury detection models within an Intensive Care Unit (ICU). This proposed framework aims to address the issues that previous MCDM approaches have encountered, including imbalanced data prioritization, ranking instability, and distortions resulting from data transformation. To address these issues, this study proposes a novel method of assigning weights, referred to as the Relational Vector Method (RVM), which utilizes expert opinion to overcome the inconsistencies inherent in the traditional pairwise comparison method. The alternatives are ranked with the Root Assessment Method (RAM). This technique provides stable, para-free, elitist, and transformation-free ranks that are independent of weighted value adjustments. It includes both benefit and cost criteria without requiring normalization or complex recalibration. A total of 32 hybrid models are evaluated based on seven evaluation criteria. Weight coefficients for these metrics are determined using a new subjective weighting technique, the RVM. RAM is applied to benchmark models and selects the best for trauma patients in intensive care units (ICUs). The RVM weighting process assigned the highest importance to classification accuracy (0.1945), precision (0.1676), and recall (0.1468). The RAM ranking identified the Gini Index with Decision Tree (GI-DT), Information Gain with Decision Tree (IG-DT), and Information Gain with Naïve Bayes (IG-Naive B) as the top-performing models with RAM scores of 0.024996, 0.024846, and 0.022041, respectively. The proposed MCDM framework is thoroughly validated using sensitivity assessment. Based on the comparative analysis performed, the results of the proposed framework outperformed the existing frameworks. This research advances the field of medical informatics by providing a methodological framework for enhancing the detection and prioritization of trauma injury models designed explicitly for ICU settings. This research contributes to the advancement of MCDM theory by addressing rank reversal, inconsistency, parameter sensitivity, and interpretability, which remain open challenges in existing methods.