Resilient and Sustainable Closed-Loop Supply Chain Design in Pharmaceuticals: A Hybrid Optimization Framework Integrating LSTM Forecasting

Supply chains in the pharmaceutical sector face increasing challenges related to demand uncertainty, product expiration, and environmental pressures, requiring robust resilient and sustainable closed-loop supply chain (CLSC) designs. This study proposes a multi-objective optimization framework for pharmaceutical CLSC design integrated with a shared pharmacy-level information system and machine learning-based demand forecasting. Demand forecasting is performed using Long Short-Term Memory (LSTM) and Recurrent Neural Network (RNN) models trained on historical time-series demand data. The dataset is split into training and testing sets, and model performance is evaluated using standard regression metrics, including Mean Squared Error (MSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). In this study, LSTM achieving coefficients of determination (R²) of 0.941 and 0.937, respectively, with corresponding MSE values of 0.051 and 0.063, indicating strong predictive capability. These forecasts are then incorporated into the optimization model as stochastic demand inputs. The optimization model simultaneously minimizes total cost and environmental impact while maximizing social responsibility in a multi-period, multi-product closed-loop network. The model is solved using the NSGA-II metaheuristic algorithm and compared with exact methods for small-scale instances. Results show that exact methods become computationally infeasible beyond problem size 10, whereas NSGA-II efficiently generates Pareto-optimal solutions for larger instances. Numerical analysis demonstrates that integrating deep learning-based demand forecasts leads to a reduction in total cost of approximately 1.5–2%, a decrease in environmental impact of 2–3%, and an improvement in social responsibility indicators of 3–5% compared to models without forecasting integration. Sensitivity analysis further indicates that increasing disruption probabilities can increase total cost by up to 37%, highlighting the importance of resilience modeling under uncertainty. The findings confirm that coupling machine learning-based forecasting with multi-objective optimization improves decision quality and robustness in pharmaceutical supply chain design.