In this paper, we propose a hybrid approach for the optimal allocation of hospital beds to emergency patients, tackling the common challenge of unlabeled clinical data. Instead of relying on predefined labels, we use a rule-based labeling step to assign initial risk levels based on key medical features. These labels are then used to train several supervised learning models, helping to improve risk prediction and explainability. The predicted levels: Minimum, Moderate, or Maximum, serve as input to an optimization model that assigns patients to beds while reducing transfer cost and supporting infection control. Results show promising improvements in fairness and allocation efficiency, with a comparative study highlighting the benefits of our method.

This study addresses the critical challenge of hospital bed assignment during emergencies by introducing a two-stage combining unsupervised learning and optimization. In the first stage, clustering techniques (K-means, GMM, DBSCAN, and HDBSCAN) are employed to automatically categorize patients into different risk levels based on clinical features, eliminating the need for labeled training data. In the second stage, a linear programming model optimizes the assignment of patients to hospital beds by minimizing costs while respecting capacity limits and medical priority constraints. The proposed approach is tested and validated on instances derived from real-world datasets from Tunisian hospitals. Our proposal approach significantly reduces the data preprocessing workload while ensuring effective prioritization of critical cases across both small and large-scale scenarios. Additionally, the optimal number of clusters is determined through silhouette analysis, enhancing the clinical relevance of the patient clutering.

ABSTRACT Rising global dependence on cloud services has become crucial for enterprises, aiming to guarantee continuous data accessibility while pursuing enhanced energy efficiency and minimized carbon emissions from data centers. However, the persistent challenge of high‐energy consumption in these facilities necessitates a concentrated approach toward energy reduction. This paper introduces an innovative multi‐objective scheduling strategy for scientific workflows, tailored for heterogeneous computing environments. Our method employs a hybrid genetic algorithm, incorporating Hill Climbing to generate an initial population of chromosomes. Subsequently, a genetic algorithm optimizes task assignments to the most suitable virtual machines, utilizing a meticulously designed fitness function to evaluate each chromosome's suitability for solving the scheduling problem. Through extensive experimentation, we demonstrate that our proposed algorithm outperforms other scheduling techniques in terms of solution quality, contributing to reduced energy consumption, processing duration, and cost. We contend that this innovative approach holds substantial potential in mitigating the energy consumption and carbon footprint associated with cloud data centers, offering a sustainable and environmentally conscious solution for scientific workflow scheduling.

The patient bed assignment problem (PBAP) consists of assigning a set of patients to a set of beds over a horizon time, by considering not only patient pathologies and state conditions, but also bed availability, and medical needs. The PBAP becomes more complex when dealing with elective patients who require specific clinical needs. A solution to this problem is not trivial to find due to the multiple constraints that are considered. The problem is hence modeled as an optimization problem that aims to minimize a set of penalties related to the non-satisfied constraints. To deal with this complex problem, this paper proposes a constraint satisfaction-based approach to solve the problem, namely the min-conflict algorithm (MCA). The proposed approach is compared with different existing solution approaches to evaluate its performance. The experiments show that the proposed MCA performs better in most benchmark instances in terms of costs and computational time.

This study focuses on the issue of assigning hospital beds to patients based on their medical condition within a specific length of stay. To handle this problem a two-stage approach combines machine learning (ML) techniques and optimization is proposed. The first level seeks to classify patients into three categories based on their contagious states by considering various features. The second one aims to minimize the cost of assigning patients to specific beds according to their states under a set of constraints (capacity constraints, transfer constraints, etc). The problem is formulated as a linear programming to increase the patient admission rate during emergencies. The proposed methodology is tested using a set of instances based on real data taken from Tunisian Hospitals. Experimental results illustrate the effectiveness of this approach, the best model is chosen based on the highest accuracy, specificity, and sensitivity.

We address, in this paper, a very recurring problem within hospitals that consists in assigning elective patients to a limited number of beds. Especially when dealing with patients requiring urgent intervention, this problem becomes more complex and the time factor becomes the most critical. In such situations, a set of patients are to be examined and their clinical states are to be well specified in order to decide whether they need admission and hospitalization or not. In case of hospitalization, the hospital staff should assign patients to beds while taking into account beds availability in terms of specialization and patient needs. All these actions should be well planned in order to maximize the quality of service in the hospitals. This challenging problem can be modeled as an assignment problem that handles a set of patients to be assigned to a set of beds over a given time horizon, while taking into account availability constraints expressed in terms of beds, medical necessity and patient demands. Due to its NP-hardness, the problem is mainly solved using approximate approaches, especially for large-scaled instances. We propose a hybrid simulated annealing approach, combining both advantages of simulated annealing (SA), that provides a local search, and genetic algorithm, that provides a global search, to enhance the performance of SA. The experimental results show that the proposed metaheuristic generates high-quality solutions for several benchmark instances from the literature with regards to the basic simulated annealing approach.

Optimisation in changing environments is a challenging research topic since many real-world problems are inherently dynamic. Inspired by the natural evolution process, evolutionary algorithms (EAs) are among the most successful and promising approaches that have addressed dynamic optimisation problems. However, managing the exploration/exploitation trade-off in EAs is still a prevalent issue, and this is due to the difficulties associated with the control and measurement of such a behaviour. The proposal of this paper is to achieve a balance between exploration and exploitation in an explicit manner. The idea is to use two equally sized populations: the first one performs exploration while the second one is responsible for exploitation. These tasks are alternated from one generation to the next one in a regular pattern, so as to obtain a balanced search engine. Besides, we reinforce the ability of our algorithm to quickly adapt after cnhanges by means of a memory of past solutions. Such a combination aims to restrain the premature convergence, to broaden the search area, and to speed up the optimisation. We show through computational experiments, and based on a series of dynamic problems and many performance measures, that our approach improves the performance of EAs and outperforms competing algorithms.

We address in this paper a special case of the online knapsack problem (OKP) that considers a number of items arriving sequentially over time without any prior information about their features. As items features are not known in advance but revealed at their arrival, we allow the decision maker to delay his decision about incoming items (to select the current item or reject it) until observing the next ones. The main objective in this problem is to load the best subset of items that maximizes the expected value of the total reward without exceeding the knapsack capacity. The selection process can be stopped before observing all items if the capacity constraint is exhausted or when the decision maker estimates that no suitable items will be received in the future. We propose to solve this problem an exact solution approach that decomposes the original problem dynamically and incorporates an optimal stopping rule in order to decide whether to load or not each new incoming item. We illustrate the proposed approach by numerical experimentations and compare the obtained results for different utility functions using performance measures. We prove therefore that the solution depends mainly on the decision maker's utility function and his readiness to take risks.