Application of Hybrid Deep Learning and Fuzzy Inference Models in Supply Chain Risk Management and Demand Forecasting: A Systematic Review

This systematic review synthesizes ۶۲ peer-reviewed studies (۲۰۰۷–۲۰۲۵) on the application of hybrid deep learning–fuzzy inference models in supply chain management, with particular emphasis on demand forecasting and the critically underrepresented yet rapidly emerging domain of supply chain risk management and resilience. The dominant paradigm—integration of Adaptive Neuro-Fuzzy Inference Systems (ANFIS) with Long Short-Term Memory (LSTM) networks and their advanced variants (CNN-LSTM-ANFIS, Attention-enhanced LSTM-ANFIS, Transformer-fuzzy hybrids)—consistently achieves ۱۵–۴۵% superior accuracy over standalone deep learning, statistical benchmarks, and pure neuro-fuzzy systems in volatile, non-stationary, and uncertain environments. While demand forecasting remains the primary application (≈۸۵% of studies), risk management applications have gained significant momentum post-۲۰۲۰, exploiting the unique neuro-symbolic complementarity of fuzzy inference: its irreplaceable capacity to model epistemic and linguistic uncertainty, qualitative risk factors, expert judgment, and tail-risk phenomena that deep learning alone systematically fails to capture. This synergy markedly enhances proactive resilience and sensitivity to extreme disruptions. Critical analysis reveals persistent gaps: over-dependence on small/synthetic datasets, scarcity of genuine multi-echelon implementations, limited real-world deployment evidence, and near-total absence of digital-twin integration. The field has reached adolescence—proven superiority in controlled settings—but remains far from industrial maturity in delivering system-level resilience impact against ۲۰۲۵-era geopolitical, climate, cyber, and regulatory shocks. Strategic advances in explainability, large-scale empirical validation, and digital-twin-coupled architectures are urgently required to realize the transformative potential of these neuro-symbolic hybrids in contemporary global supply chains.

کلیدواژه ها:

ANFIS-LSTM hybrid models ، Epistemic uncertainty ، Neuro-symbolic artificial intelligence ، Supply chain resilience ، Supply chain risk management ، Tail-risk sensitivity