This study investigates the value of incorporating diversification as an explicit objective in sustainable portfolio optimization, alongside expected return, risk measured by Value-at-Risk (VaR), and sustainability performance. Since minimizing VaR and maximizing sustainability do not necessarily lead to well-diversified portfolios, we assess the incremental contribution of diversification, measured by the Diversification Ratio (DR). To this end, a four-objective Mean–VaR–ESG-diversification model is compared with a reduced three-objective Mean–VaR–ESG formulation. Given the non-convex nature of the resulting optimization problems, both formulations are solved using an adapted version of ev-MOGA. The study further incorporates investor preferences through a preference-direction framework. Instead of requiring predefined objective weights, this approach allows investors to specify their preferred direction in the objective space, facilitating the identification of portfolios that reflect different priorities across return, downside risk, and sustainability. The proposed methodology is evaluated through an extensive out-of-sample analysis of four equity datasets based on major stock indices: the Euro Stoxx 50, FTSE 100, S&P 100, and Nasdaq 100. The results show consistent improvements in diversification, volatility, and VaR, with generally modest reductions in ESG scores and trade-offs in return performance that vary across datasets and investor profiles.
Epidemic resource-allocation decisions combine rapidly changing demand, incomplete reporting, and conflicting evidence. We formulate an Ambiguous Cognitive Map (ACM) in which each concept has four coupled evidence channels—true, false, partially true, and partially false. The evidence-conditioned operator maps to a capped convex state space, admits a verifiable contraction condition, and generates priorities for a constrained allocation layer. Evaluation combines 30 paired synthetic supply-chain runs, observational NHS England and U.S. Department of Health and Human Services hospital panels, retrospective GetUsPPE assignment records, and PPE-Match request, offer, and distance streams. Hyperparameters for the two allocation replays are selected by seeded tree-structured Parzen-estimator optimization on temporally separated calibration and validation periods, with the final periods held out. ACM attains NDCG@3 values of 0.948 and 0.905 on the NHS holdout and external HHS panels, respectively, while persistence-based comparators rank subsequent stress more accurately. On held-out GetUsPPE assignments, calibrated ACM reaches an average precision of 0.081 and allocation overlap of 0.008, compared with 0.177 and 0.110 for a scalar baseline. On held-out PPE-Match streams, calibration raises the prespecified operational composite from 0.132 to 0.200 and reduces unit-miles from 1999.6 to 24.8; ACM exceeds demand-only allocation (0.161) but remains below equal priority (0.204). Coupled and independent four-channel variants are operationally indistinguishable in both replays. ACM consequently provides a mathematically bounded and auditable ambiguity-aware decision interface, with performance that depends on the operational endpoint and data environment.
Post-consumer waste management has been a growing concern regarding manufacturing, especially with the expansion of the plastic industry. Decision-making in such settings often relies on a range of factors in different aspects, rather than forecasting accuracy and sustainability. Although time series forecasting and Multi-Criteria Decision Making (MCDM) in sustainable manufacturing have been widely studied, their integration is still limited, particularly when uncertainty in forecasting is considered. MCDM methods often do not incorporate forecasting results, limiting their ability to simulate realistic environments, whereas most traditional forecasting models are built under Gaussian assumptions and mainly focus on point predictions, restricting dynamic uncertainty estimation. On the other hand, recent approaches attempt to simulate uncertainty but often come with increased complexity not only in requirements but also in model comprehension, limiting their applicability. This study proposes an integrated framework combining probabilistic forecasting with MCDM to address such issues while maintaining sustainability in manufacturing. Additionally, a bootstrapping technique is applied to construct prediction intervals, discarding any specific distribution assumption, which elevates the potential of classical models. The case study results demonstrate that the proposed hybrid framework outperforms traditional and probabilistic approaches by 6% and 3% in forecasting accuracy, respectively, while simultaneously facilitating more balanced decision-making across environmental, economic, technical, and social dimensions, thereby enhancing the overall sustainability and practicality of manufacturing operations.
Capacitated Facility Location Problems under uncertainty are of great significance in real-world location scenarios. We propose a Capacitated Facility Location Problem under multiple uncertainties (CFLP-MU) that considers demand fluctuations, facility disruption risks, and capacity availability uncertainty. The model incorporates flexible capacity expansion and reactive recovery mechanisms. To solve this problem, we aim to minimize total costs while maximizing the expected Service Satisfaction Ratio (SSR). We develop a DSAA-Hybrid NSGA-II framework that outperforms baseline MOEAs (e.g., NSGA-II, MOEA/D, SPEA2) in both convergence efficiency and Pareto quality under uncertainty. This framework is driven by four innovative mechanisms: a Deterministic Warm Start (DWS), a Dynamic Sample Average Approximation (DSAA), a VIP capacity reservation policy, and a Feedback-driven Smart Repair (FSR) operator. Extensive numerical experiments reveal that capacity expansion with the tiered pricing structure reduces the need for idle backup capacity by buffering demand surges at a lower fixed cost than physical redundancy. Sensitivity analysis of medium scale demonstrates that stochastic planning yields superior value under high-penalty operational constraints, while the marginal utility of perfect information diminishes as network redundancy increases. This research offers a practical tool for balancing economic efficiency and reliability in strategic network design under uncertainty.
Rank aggregation problem is fundamental to multi-criteria decision analysis and model prioritization in logistics planning. In particular, the Kemeny framework achieves a consensus ranking by minimizing the Kemeny distance to the input rankings. Although many heuristics exist for the NP-hard Kemeny problem, these algorithms suffer from scalability issues, especially when ties are allowed. To address these limitations, we propose two complementary heuristics, BridgeSubiteration and GreedyReposition. BridgeSubiteration constructs a consensus ranking by aggregating smaller subsets of objects, while allowing ties via a Bridging mechanism. GreedyReposition makes locally optimal decisions using a proposed repositioning operator. This operator generates a set of neighbor rankings in the tied ranking space. Through real and simulated datasets, we demonstrate that the proposed algorithms achieve lower Kemeny distance in comparable computational time, making large-scale Kemeny Rank Aggregation practically feasible.
