Applied Soft Computing最新文献

{"title":"Cost investigations and ANFIS implementation for M/M/1/L queueing model with general vacation, customer discouragement, and F-policy","authors":"Khushbu S. Antala,&nbsp;Sudeep Singh Sanga","doi":"10.1016/j.asoc.2026.114667","DOIUrl":"10.1016/j.asoc.2026.114667","url":null,"abstract":"<div><div>This work investigates the M/M/1/L queueing model within real-life scenarios, highlighting its substantial impact on various domains such as call centers, distributed computing systems, healthcare facilities, traffic management centers, and other service-oriented operations. It focuses on the finite capacity queueing model with general vacations and customer discouragement under the <em>F</em>-policy. Additionally, we consider essential queueing features that account for customers’ behaviors, such as balking and reneging. When customers encounter long queues, they may experience discouragement and consequently choose not to join the queue or leave without receiving service. Implementation of the ‘<em>F</em>-policy’ is an effective strategy for the admission of customers into the system to reduce congestion resulting from excessive customer arrivals. Further, the server chooses to take a vacation when no customer is present in the system. The vacation time is not exponentially distributed; instead, we consider the general case of vacation time for a more comprehensive analysis. The mathematical development of the M/M/1/L queueing model is carried out using the Chapman-Kolmogorov steady-state equations by introducing supplementary variables corresponding to remaining vacation times. Subsequently, the Laplace-Stieltjes transform and recursive method are employed to solve these equations and establish probability distributions. Further, various performance metrics, such as the number of customers in the system, throughput, customer loss, long-run probabilities, etc., are derived. These performance measures will assist system organizers in making informed decision-making strategies. A numerical example is also presented, illustrating the impact of input parameters and customers’ discouraged behavior on various performance metrics. The adaptive neuro-fuzzy inference system, which is built on an artificial neural network and a support vector regression, a machine learning technique, is employed to validate the numerical results. Moreover, the nonlinear cost function is formulated with the service and vacation rates as decision variables. The cost is minimized using quasi-Newton methods and several metaheuristics (particle swarm optimization, bat algorithm, and differential evolution variants). These algorithms are utilized to compare the optimal values of the cost function. The queueing model’s practical application is demonstrated through its implementation in fog computing systems.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114667"},"PeriodicalIF":6.6,"publicationDate":"2026-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080042","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"LightPro: Lightweight multi-channel model protection for edge-oriented deep neural network","authors":"Jinyin Chen ,&nbsp;Tianxin Zhang ,&nbsp;Xiaoming Zhao ,&nbsp;Zekai Wang ,&nbsp;Haibin Zheng","doi":"10.1016/j.asoc.2026.114605","DOIUrl":"10.1016/j.asoc.2026.114605","url":null,"abstract":"<div><div>It is well known that deep neural networks (DNNs) rely on large-scale proprietary data and computing-intensive resources to train, high-quality DNNs have become one of the most valuable assets in the supply chain. This also attracts attention of the model stealers for great profits, as they can steal functionally equivalent models by iteratively collecting outputs using carefully crafted examples. Extensive studies on model protection have been carried out against stealing attacks, nevertheless, they are still challenged in aspects of effectiveness, efficiency, and transferability. To this end, we propose <em>LightPro</em>, a lightweight multi-channel authorization cryptographic protection method for DNNs. LightPro significantly differs from previous work in four aspects, <em>i.e.,</em> i) <em>lightweight</em>: its encryption process only processes the input and output layers without complex processing of inner information, <em>e.g.,</em> model structure and parameters; ii) <em>robust</em>: it provides defense against various reverse attacks by embedding an additional key-verified encryption channel in the input layer and a dynamic noise layer in the output layer; iii) <em>good trade-off</em>: it introduces a noise layer in the output layer to blur the encryption method of the probability score distribution without changing the score ranking thereby maintaining high performance on primary tasks with negligible accuracy degradation; iv) <em>transferable</em>: its flexibly mappable encrypted channel added at the input layer can adapt to different types of inputs, thereby facilitating the application of different tasks, data, and models. Extensive experiments on 8 datasets, and 6 tasks, <em>i.e.,</em> image classification (AlexNet, VGG16, GoogLeNet), object detection (SSD, Faster R-CNN, YOLOX), text classification (BERT), speech classification (Deepspeech), signal classification (ResNet) and graph classification (GCN), compared to five baselines against two attacks testify that LightPro achieves the state-of-the-art (SOTA) performance in terms of effectiveness, efficiency, generality and extensibility. For example, it significantly reduced the accuracy of the attack model by 73.02% on average compared to the baselines. Its time cost is reduced to <span><math><mo>∼</mo></math></span> 1/49 of the baselines, and the mean average precision of the attack model is decreased by 4.21<span><math><mo>∼</mo></math></span>5.88 times on average.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114605"},"PeriodicalIF":6.6,"publicationDate":"2026-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039724","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Fuzzy concept-cognitive learning based on attention mechanism and three-way partial order structure","authors":"Lufang Zhang ,&nbsp;Enliang Yan ,&nbsp;Tianyong Hao","doi":"10.1016/j.asoc.2026.114664","DOIUrl":"10.1016/j.asoc.2026.114664","url":null,"abstract":"<div><div>Concept cognition plays an important role in the implementation of cognitive intelligence, while the attention mechanism is crucial in human cognition. Fuzzy concept-cognitive learning (CCL) based on the attention mechanism represents a promising research direction. Partial order formal structure analysis (POFSA) is an emerging CCL model. However, its reliance on binary formal contexts makes it difficult to handle fuzzy data, which limits its practical usage since many real-world phenomena are inherently fuzzy and cannot be accurately described using binary logic. This article aims to address the gap in fuzzy CCL within the POFSA domain. We propose a fuzzy CCL method based on the attention mechanism and three-way partial order structure. Specifically: 1) Some fuzzy concept granules are defined based on the fuzzy formal decision context; 2) By introducing the attention mechanism to explore the characteristics of fuzzy data and drawing on the idea of a partial order three-way structure, a fuzzy partial order three-way structure based on the attention mechanism is constructed; 3) Classification experiments are conducted to verify that the model can be applied to other fields. Experiments on 11 datasets show that the proposed method can effectively distinguish new concept categories and has excellent classification performance.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114664"},"PeriodicalIF":6.6,"publicationDate":"2026-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039883","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Novel mathematical modeling of IT2 Takagi–Sugeno fuzzy PID controllers with triangular FoUs and their application: A MagLev case study","authors":"Debdoot Sain,&nbsp;Tae H. Lee","doi":"10.1016/j.asoc.2026.114663","DOIUrl":"10.1016/j.asoc.2026.114663","url":null,"abstract":"<div><div>This study presents two new mathematical models of interval type-2 Takagi–Sugeno fuzzy proportional-integral-derivative (IT2TSFPID) controllers, incorporating triangular footprints of uncertainty (FoUs), specifically the bottom-wide triangular (BWT) and top-wide triangular (TWT) structures. Unlike previous one-dimensional (1D) input-space-based studies that predominantly employed parallelogram-shaped FoUs to model IT2TSFPID controllers, this study is the first to utilize triangular FoUs in their design. The proposed models compute the scaled incremental control effort by algebraically summing the individual scaled incremental contributions from the proportional, integral, and derivative components, each modeled within the interval type-2 (IT2) Takagi–Sugeno (TS) fuzzy framework. A distinctive feature of this work is the use of the modulus of scaled input variables in rule consequents, which reduces the number of tunable parameters and overall modeling complexity. Furthermore, unlike the existing triangular FoU-based IT2 Mamdani fuzzy proportional-integral-derivative (IT2MFPID) controllers, the absence of bias terms in the controller expressions ensures better adherence to the nonlinear PID structure. Notably, the proposed controllers are developed using the centroid defuzzifier, a method not previously applied to IT2TSFPID models. The use of 1D input spaces further allows independent tuning of IT2TSFPID gains, offering improved flexibility in controller design. In addition to modeling, this study provides a detailed analysis of the properties, including bounded-input bounded-output (BIBO) stability, of the proposed controllers, aimed at delivering a comprehensive understanding of their behavior. Finally, rigorous simulation results from a magnetic levitation (MagLev) case study are presented to demonstrate the effectiveness and practical relevance of the proposed controllers, along with extensive comparisons made against the existing work.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114663"},"PeriodicalIF":6.6,"publicationDate":"2026-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080109","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Non-exemplar class-incremental learning: Dynamic adversarial sample synthesis and relational knowledge distillation","authors":"Yuanlong Yu,&nbsp;Xiaoli Ke","doi":"10.1016/j.asoc.2026.114601","DOIUrl":"10.1016/j.asoc.2026.114601","url":null,"abstract":"<div><div>Deep neural networks exhibit robust performance on isolated tasks but are prone to catastrophic forgetting when learning sequential tasks. While various incremental learning paradigms have been developed to address this, they commonly rely on stored old data or generative models, which introduce significant limitations. Storing past data raises privacy and memory concerns, whereas generative models often face training instability and high computational costs. To overcome these challenges, this paper proposes a novel non-exemplar incremental learning framework, termed MRDAS, designed to effectively alleviate catastrophic forgetting without retaining any previous data. Our framework integrates two key mechanisms: (1) Dynamic Adversarial Sample Synthesis (DASS), which generates pseudo-features for old classes to balance the learning between old and new classes; and (2) Multi-layer Relational Knowledge Distillation (MRKD), which transfers inter-sample relational knowledge across multiple network layers to preserve structural information. Extensive experiments on CIFAR-100, ImageNet-Subset, and Tiny-ImageNet demonstrate the superiority of MRDAS. For example, on the 10-phase CIFAR-100 benchmark, MRDAS achieves a final average accuracy of 68.93%, outperforming the strongest baseline by 3.46 percentage points. It also reduces the forgetting rate to 8.85%, which is 7.29% lower than the best comparable method. The proposed approach thus offers a privacy-preserving, stable, and efficient solution for continual learning, providing a practical pathway toward deployable lifelong learning systems that require neither data rehearsal nor complex generative models.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114601"},"PeriodicalIF":6.6,"publicationDate":"2026-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039858","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"RADSCL: Representation augmentation integrated with dual supervised contrastive learning for low-resource text classification","authors":"Jun Zhang ,&nbsp;Ze Kuang ,&nbsp;Fanfan Shen ,&nbsp;Jianfeng He ,&nbsp;Rui Sun ,&nbsp;Yanxiang He","doi":"10.1016/j.asoc.2026.114646","DOIUrl":"10.1016/j.asoc.2026.114646","url":null,"abstract":"<div><div>Text classification in low-resource regime is a challenging task. Data augmentation techniques can significantly alleviate the issue of insufficient training samples in such environments by generating new samples. However, existing data augmentation methods have not yet effectively solved the problems of hard samples that are hard to classify and insufficient model generalization ability, which makes the performance of text classification in low-resource regime still has room for improvement. To this end, this paper proposes a method that fuses representational data augmentation and dual supervised contrast learning (RADSCL) in low-resource regime. Representational data augmentation first uses dynamic span-cutoff to remove context-independent words, which reduces the parameters needed for mixup and lowers the computational cost. Then the meaningless property of PAD word embeddings is utilized to perform weighted mixing with the cut-off text to reduce the importance of certain words in the text that affect classification. Through the representation augmentation method proposed in this paper, high-quality hard positive samples can be generated to optimize the decision boundary of the model. On this basis, in view of the current contrastive learning has achieved significant results in enhancing text representation, this paper constructs a dual supervised contrastive learning framework. The framework not only uses supervised contrastive learning to focus on the contrastive relationship between positive and negative samples, but also fully learns the potential semantic relationship between different categories of samples under the role of soft labels. Label distribution contrastive learning further utilizes the distribution information of soft labels based on supervised contrastive learning to impose effective constraints on the text representation, which effectively improves the performance of the text classification model. Multiple sets of experimental results show that the performance of the RADSCL model outperforms any model that incorporates other data augmentation and supervised contrastive learning models by a considerable margin on three benchmark datasets. In particular, the average accuracy of RADSCL is improved by 2.76 %, 2.66 %, 0.91 %, and 2.75 %, respectively, relative to the fusion models such as EDA+CE+SCL, Mixup+CE+SCL, ADV+CE+SCL, and AWD+CE+SCL.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114646"},"PeriodicalIF":6.6,"publicationDate":"2026-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039870","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Data-driven machine learning surrogate models for estimating the wind response of tall buildings","authors":"Magdy Alanani ,&nbsp;Marwa Allam ,&nbsp;Ahmed Elshaer","doi":"10.1016/j.asoc.2026.114638","DOIUrl":"10.1016/j.asoc.2026.114638","url":null,"abstract":"<div><div>As the demand for high-performance structures continues to rise, the need for performance-based wind design (PBWD) methodologies becomes increasingly apparent. The utmost outputs of PBWD can be achieved when adopted with an optimization framework. However, the computational costs are prohibitively high for optimizing tall building layouts using high-fidelity simulations. Thus, integrating surrogate models offers a promising solution, enabling computationally affordable design optimization. This paper presents a comprehensive assessment of different data-driven models to develop a surrogate model of tall buildings subjected to dynamic wind loads based on their structural system layout. The study investigates the performance of six different machine learning models: Ridge Regression (RDG), Decision Trees (DT), Random Forests (RF), Extreme Gradient Boosting (XGB), Support Vector Machines (SVM), and Deep Neural Networks (DNN). A training and testing dataset has been prepared using Finite Element Method (FEM) for a case study building subjected to wind load time history generated by an experimentally validated Computational Fluid Dynamics (CFD) model. For a valid comparison, each developed surrogate model is trained using a grid search algorithm through a k-fold cross-validation process to achieve optimal parameters and hyperparameters of each model. The investigation involves a detailed analysis of statistical performance metrics applied to training and testing datasets. The results show that the DNN model achieves the highest predictive accuracy, reaching an <span><math><msup><mrow><mi>R</mi></mrow><mrow><mn>2</mn></mrow></msup></math></span> of 0.84 for demand-to-capacity ratio prediction under single-angle wind loading, while XGB provides competitive performance with faster inference times. The proposed surrogate framework enables rapid, data-driven estimation of global responses (peak displacement and drift) and local component demands (wall and column D/C ratios), providing a computationally efficient pathway for integrating optimization, sensitivity analysis, and early-stage PBWD decision-making.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114638"},"PeriodicalIF":6.6,"publicationDate":"2026-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039867","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"MHADFormer: A cost-efficient multiscale hybrid transformer mixer model for automating Alzheimer’s disease diagnosis from MRI scans","authors":"Francis Jesmar P. Montalbo","doi":"10.1016/j.asoc.2026.114624","DOIUrl":"10.1016/j.asoc.2026.114624","url":null,"abstract":"<div><div>Accurate and early diagnosis of Alzheimer’s disease (AD) from magnetic resonance imaging (MRI) remains a critical challenge in clinical neuroscience. Deep learning and vision transformers have advanced automated classification, yet their high computational burden limits deployment in practice. Here, a hybrid architecture, Mixer Hybrid AD Transformer (MHADFormer), is introduced to reconcile accuracy with efficiency through four lightweight modules: the Cost-efficient Feature Extractor (CeFE), Enhanced Mobile Vision Transformer (EMViT), Fused Attention with Cost-efficient Separable Convolutions (FACeS), and Feature Aggregated Spectral Transformer (FAST), integrated with a tailored MLP-Mixer. MHADFormer achieved 99.69 % accuracy on the OASIS dataset and 99.96 % on ADNI, averaging 99.83 %, while requiring only 1.11 M parameters and 0.27 GFLOPs. In addition, the model achieved 81.28 % accuracy in distinguishing between progressive and stable mild cognitive impairment, surpassing state-of-the-art baselines. These results establish MHADFormer as a scalable, interpretable, and resource-aware solution, bridging the gap between deep learning performance and clinical applicability for MRI-based AD diagnosis.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"190 ","pages":"Article 114624"},"PeriodicalIF":6.6,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980375","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-level retrieval with representation alignment enhances cross-document evidence synthesis for scientific knowledge generation","authors":"Shenghua Wang ,&nbsp;Zhongqiang Wang ,&nbsp;Chunhe Qu ,&nbsp;Zhen Yin","doi":"10.1016/j.asoc.2026.114649","DOIUrl":"10.1016/j.asoc.2026.114649","url":null,"abstract":"<div><div>Recent advancements in retrieval-augmented generation (RAG) have significantly improved performance in knowledge-intensive natural language processing tasks. However, applying RAG to scientific literature for knowledge synthesis remains challenging due to the difficulty in effectively retrieving and integrating multi-element evidence across documents and in modeling long-range dependencies. Moreover, the absence of dedicated evaluation resources for multi-element, cross-document scientific RAG further limits progress in this area. To address these challenges, this paper proposes a novel approach that integrates multi-level retrieval and contrastive learning to enhance RAG for scientific literature. Specifically, contrastive learning is employed to optimize embeddings for retrieving diverse element types, including text, tables, formulas, and figures, while a multi-level retrieval strategy combines coarse-grained document filtering with fine-grained element ranking to improve retrieval accuracy and efficiency. In addition, Chain-of-Thought (CoT) reasoning is incorporated to enhance the coherence of synthesized knowledge. To support research in this direction, we construct SciLitQA, a new dataset comprising over 50,000 high-quality Q&amp;A pairs tailored for evaluating multi-element, cross-document knowledge synthesis. Extensive experiments on SciLitQA and public benchmarks demonstrate the effectiveness of our approach and establish SciLitQA as a valuable resource for advancing research in scientific RAG.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114649"},"PeriodicalIF":6.6,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039912","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A hybrid NSGA-II and delayed local search approach for home healthcare scheduling and routing optimization","authors":"Radhia Zaghdoud ,&nbsp;Issam Zidi ,&nbsp;Olfa Ben Rhaiem ,&nbsp;Salim El Khediri ,&nbsp;Khaled Mesghouni","doi":"10.1016/j.asoc.2026.114617","DOIUrl":"10.1016/j.asoc.2026.114617","url":null,"abstract":"<div><div>The home healthcare sector has emerged as a vital component of modern healthcare systems, addressing the needs of aging populations and individuals with disabilities. However, efficiently managing caregiver schedules and routing while balancing service time windows, skill compatibility, and operational costs remains a critical challenge. This study tackles a multi-objective Home Healthcare Scheduling and Routing Problem (HHCSRP) by optimizing three competing objectives: minimizing total travel time, ensuring equitable workload distribution among caregivers, and reducing waiting time between patient visits. We propose a novel two-stage hybrid algorithm that strategically integrates the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) with a tailored local search method. In the first stage, NSGA-II generates a diverse initial population over multiple generations, while the second stage embeds local search at mid-generations to refine solutions without premature convergence. Computational experiments using adapted Solomon’s Vehicle Routing Problem with Time Windows (VRPTW) benchmarks demonstrate the effectiveness of the algorithm. Our approach achieves superior hypervolume (0.57–0.86) and Pareto front diversity (<span><math><mo>|</mo><mi>F</mi><mi>P</mi><mo>|</mo></math></span> = 94.27 on average), outperforming existing methods in balancing efficiency, fairness, and scalability. The results highlight the robustness of the method in handling large-scale instances (100 patients), offering a scalable tool for real-world home healthcare logistics. This work advances multi-objective optimization in healthcare operations, providing actionable insights for administrators to harmonize patient satisfaction and operational efficiency.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114617"},"PeriodicalIF":6.6,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039739","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}