Purpose: Quality management in pharmaceutical supply chains requires the simultaneous evaluation of conflicting criteria, interdependencies among criteria, and uncertainty in decision-makers’ judgments. This study aims to develop an adaptive framework for pharmaceutical supplier selection that accounts for interdependencies among criteria and updates criterion weights based on performance feedback.

Methodology: This study proposes the DG-INFAO multi-criteria decision-making framework, which integrates graph-based aggregation with uncertainty-aware feedback learning. Within this framework, interval-valued trapezoidal neutrosophic numbers are used to represent uncertain evaluations, a graph structure is employed to model interdependencies among criteria, and an adaptive learning mechanism is used to update criterion weights. The framework is evaluated using simulated data for the selection of six pharmaceutical suppliers through convergence analysis, comparison with benchmark methods, ranking stability assessment, and sensitivity analysis.

Findings: The results indicate that the algorithm converges after 358 iterations, with the root mean square error (RMSE) decreasing from 0.067405 to 0.066898. The third and fifth suppliers rank first and second, respectively. Sensitivity analyses involving the graph aggregation parameter and random perturbations of the graph structure confirm the stability of these two suppliers’ rankings across the scenarios examined. However, cross-validation does not demonstrate the superiority of the proposed framework across all evaluation metrics. Therefore, further investigation is required before the findings can be generalized to real-world data.

Originality/Value: The main contribution of this study lies in integrating interval-valued neutrosophic evaluations, graph-based modeling of criterion interdependencies, and uncertainty-aware feedback learning within a unified framework. The framework’s added value is assessed through adaptive criterion-weight updating, ranking stability analysis, and sensitivity analysis of the results. Given the use of simulated data and the limited number of alternatives, the framework’s empirical validity and generalizability should be further examined in future studies using real-world data and larger decision-making sets.