DEEP FUZZY NEURAL NETWORKS FOR MULTI-CRITERIA DECISION MAKING IN UNCERTAIN ENVIRONMENTS

Abstract

Multi-criteria decision-making (MCDM) in uncertain environments faces challenges due to imprecise data, vague criteria weights, and dynamic conditions. Traditional methods like TOPSIS and AHP struggle with these complexities, while purely data-driven deep learning models lack interpretability. This paper proposes a Deep Fuzzy Neural Network (DFNN) that synergizes fuzzy logic and deep learning to enhance decision accuracy while maintaining explainability. The DFNN integrates fuzzy membership functions for uncertainty handling with deep neural networks for adaptive pattern learning, optimizing criteria weights dynamically. Evaluated on real-world datasets from healthcare, finance, and supply chain domains, the DFNN outperforms conventional methods (92.3% accuracy vs. 80.6–88.7% for baselines) and demonstrates superior robustness (RS = 0.91) and interpretability (II = 4.5/5). Ablation studies confirm the necessity of both fuzzy and deep learning components. The model bridges the gap between precise data-driven learning and transparent rule-based reasoning, making it a robust solution for high-stakes decision-making under uncertainty. Future work will explore reinforcement learning extensions and federated implementations for broader applicability.