Applied Intelligence最新文献

{"title":"A data augmentation-based deep reinforcement learning approach for high-risk states avoidance in sepsis treatment","authors":"Qiwei Zhang,&nbsp;Bin Lin,&nbsp;Yunlong Liu","doi":"10.1007/s10489-026-07393-w","DOIUrl":"10.1007/s10489-026-07393-w","url":null,"abstract":"<div><p>Sepsis is a serious infection-related condition and a leading cause of death in Intensive Care Units (ICUs), yet its optimal treatment remains uncertain. With the advancement of Deep Reinforcement Learning (DRL), it has been increasingly utilized for real-time sepsis management. However, current methods hinge on ample data to derive optimal dosing strategies while the data is difficult to obtain. Furthermore, certain treatments carry risks of organ damage and bleeding. To alleviate these issues, we propose a DRL model based on data augmentations that focuses on avoidance of high-risk states. Firstly, a medWGAN network that uses an autoencoder and the Wasserstein distance metric is used to produce more realistic synthetic data. Subsequently, two independent DRL models are developed to assess ‘dead-end’ states and predict positive treatment outcomes, enhancing patient state estimation accuracy. Finally, a strategy network with a feasible state analysis chain is proposed and incorporated into the above-mentioned framework, which can remove implausible states, and implement a dual-constraint mechanism to avoid high-risk states in treatment strategies. Experiments on the MIMIC-III dataset, which contains 17,898 sepsis patient records, demonstrate the efficacy of our approach. It generates synthetic data with distributions closely resembling real training data and accurately identifies high-risk states while optimizing treatment policies. Our approach significantly outperforms baselines and the physician’s strategy, where the Weighted Double Robust estimator (WDR)’s off-policy value is 14.28 (95% CI: [13.59, 14.87]) compared to 9.96 (95% CI: [9.57, 10.35]) for the physician’s strategy (<span>(p&lt;0.001)</span>), and the model-estimated sepsis mortality rate is 7.2%, lower than existing baselines.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 16","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176632","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 Michigan style genetic fuzzy system framework for multiple sclerosis: towards a digital twin application","authors":"Muhammad Faisal,&nbsp;Moataz Ahmed,&nbsp;Sagheer Khan,&nbsp;Reem F. Bunyan","doi":"10.1007/s10489-026-07545-y","DOIUrl":"10.1007/s10489-026-07545-y","url":null,"abstract":"<div><p>Digital Twin (DT) technology is increasingly being explored in healthcare to support personalized monitoring and decision-making. A critical requirement of any clinical DT is an interpretable and reliable predictive model that can represent the evolving condition of individual patients. This study presents a Michigan-style Genetic Fuzzy Rule-Based System (MiG-FRBS) developed as the predictive analytics core of a DT for Multiple Sclerosis (MS). The proposed framework is designed to convert sparse, irregular clinical records into meaningful, explainable forecasts of disease progression. MiG-FRBS combines fuzzy logic with genetic algorithms to automatically generate compact and human-readable rules. Unlike conventional black-box Machine Learning (ML) models, the system provides transparent reasoning by identifying key features, temporal patterns, and decision rules that describe changes in Expanded Disability Status Scale (EDSS) functional subscores. A Multiple Input Multiple Output (MIMO) structure is employed to predict several subscores simultaneously, enabling a more comprehensive virtual representation of patient status. Spline-based interpolation was used to create synthetic time-series data to allow continuous forecasting because Ms data are usually sparse and unevenly distributed throughout time. The framework accepts both raw and differenced inputs and makes use of Gaussian membership functions. With accuracies above 90% for Mental scores and above 80% for Brainstem and Bowel scores, experimental evaluation on actual MS patient data showed promising performance while preserving good interpretability through compact, succinct rule bases. In order to assist future data-driven clinical decision support for MS management, the suggested MiG-FRBS model offers an adaptable and transparent forecasting engine that can function as the virtual modelling layer of a DT.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 16","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176710","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Rough sets under incomplete information: Concept formation and three-way decision perspectives","authors":"Yiyu Yao,&nbsp;Junfang Luo,&nbsp;Qiaoyi Li","doi":"10.1007/s10489-026-07518-1","DOIUrl":"10.1007/s10489-026-07518-1","url":null,"abstract":"<div><p>A research direction in generalizing Pawlak rough sets is to consider incomplete information. One of the common practices is to replace an equivalence relation on a set of objects by various similarity relations to account for the incomplete information. Typically, one defines similarity relations based on some intuitive understanding of incomplete information. What lacks in the existing studies is a satisfactory formal formulation on a sound semantics foundation. In this paper, we present a formal model for interpreting and constructing concept-based granular rough sets under incomplete information, with a direct reference to one of the main purposes of rough set theory for data analysis and concept forming. Our formulation is based on a conjunctive concept space. For an object, we identify the most specific conjunctive concept to which the object is an instance. By using the ideas of possible world semantics of incomplete information suggested by Lipski, for the conjunctive concept, we derive a pair of extensions, namely, the set of objects that are known to be instances and the set of objects that are possible instances. We show that these two extensions are the neighborhoods induced by two similarity relations, which provides a formal and sound semantics of similarity relations. Finally, we use the proposed model to probe into the ideas of concept-based granular rough sets and three-way decision under incomplete information. Our investigation on the semantics reveals some difficulties with existing methods and calls for more studies on semantics related issues. The result will be useful in guiding future research and preventing misuse of models of incomplete information.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 16","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176375","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Robust tri-factor decomposition for multi-view clustering","authors":"Shaojun Shi,&nbsp;Jiahao Zhao,&nbsp;Yibing Liu","doi":"10.1007/s10489-026-07499-1","DOIUrl":"10.1007/s10489-026-07499-1","url":null,"abstract":"<div><p>Multi-view clustering has garnered widespread attention due to its ability to effectively mine complementary and consistent information across different views. However, traditional methods often suffer from high computational complexity, making them unsuitable for large-scale data scenarios. Although matrix factorization based multi-view clustering approaches have alleviated scalability issues to some extent, they still exhibit three main limitations: 1) Most methods fail to consider pervasive noise interference in large-scale data, resulting in significantly degraded discriminative capability and robustness in complex noisy environments; 2) They typically treat views of varying qualities equally, preventing full exploitation of the rich discriminative information in high-quality views; 3) Most methods separate clustering partition from coefficient matrices learning, preventing clustering partitions from guiding discriminative coefficient learning and thereby limiting clustering performance. To address the aforementioned issues, this paper proposes a Robust Tri-Factor Decomposition for Multi-view Clustering (RTFDMC) method. RTFDMC incorporates the <span>( l_{2,1} )</span>-norm when constructing coefficient matrices, effectively suppressing noise interference and enhancing model robustness; meanwhile, it employs a self-supervised weighting mechanism that fully leverages information from high-quality views; furthermore, RTFDMC employs tri-factor decomposition to tightly couple clustering partitioning with representation learning, thereby improving the discriminative capability of the coefficient matrices and further enhancing clustering performance. Comprehensive experiments on eight datasets validate the effectiveness of the proposed method.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 16","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176709","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Automated optimization modeling and auditing via dual-team multi-LLM-agent cooperative framework","authors":"Yuan Chiang,&nbsp;Wenhui Fan","doi":"10.1007/s10489-026-07536-z","DOIUrl":"10.1007/s10489-026-07536-z","url":null,"abstract":"<div><p>Translating natural language descriptions into solvable mathematical programs utilizing large language models (LLMs) remains a significant challenge in optimization modeling. Traditional approaches often struggle with error propagation between logical modeling and code implementation, while existing evaluation metrics frequently overlook the structural integrity of the mathematical models. This paper presents a dual-team multi-LLM-agent cooperative framework designed to automate optimization modeling and auditing through a decoupled architecture. The proposed system separates the task into a logical abstraction team that formalizes and audits the mathematical model and a physical implementation team that transcribes the audited model into solver code, with an audited mathematical model as their only interface. To improve the reliability of the generated results, we design an automated auditing protocol comprising a model auditor that jointly checks semantic fidelity and variable definitions, together with an execution-based syntax evaluator and a semantic evaluator for the generated program. Furthermore, we employ Graph Edit Distance to provide a structural assessment of model quality without requiring manual annotation on problem texts. Experimental results on the NL4Opt, ComplexOR, and MAMO datasets with (mixed-integer) linear problems demonstrate that our framework effectively reduces implementation failure and enhances the accuracy of both the mathematical formulation and the final solver code.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 16","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176631","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"3L-GBC: structure-aware granular-ball clustering for layered structure discovery","authors":"Lijie Yin,&nbsp;Yuefeng Ma,&nbsp;Lin Dai,&nbsp;Meng Wang","doi":"10.1007/s10489-026-07543-0","DOIUrl":"10.1007/s10489-026-07543-0","url":null,"abstract":"<div><p>Many synthetic and real-world datasets contain clusters with heterogeneous internal densities, where dense central regions gradually transition to sparse boundaries. However, most current clustering methods rely on single-layer modeling or rigid partitioning criteria, which limits their ability to capture complex cluster shapes and heterogeneous density distributions. To address these limitations, a density-gradient-guided three-layer granular ball clustering method (3L-GBC) is proposed. The method adopts a three-layer granular-ball representation to model dense, transitional, and sparse regions within clusters. Within this framework, candidate centers are efficiently identified using a lightweight density peak–inspired strategy, followed by the construction and adaptive expansion of density-gradient-guided granular balls. A curvature-based mechanism is further introduced to reveal an explicit three-layer internal structure. Finally, granular balls are merged into complete clusters through adaptive density propagation driven by multi-regional structural consistency. Experimental results on synthetic and real datasets demonstrate that 3L-GBC achieves competitive clustering performance on complex structures.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 16","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176373","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Dynamic attention for container relocation problems: a deep reinforcement learning approach","authors":"Zizhen Zhang,&nbsp;Fengwei Liu,&nbsp;Mengchu Zhou","doi":"10.1007/s10489-026-07525-2","DOIUrl":"10.1007/s10489-026-07525-2","url":null,"abstract":"<div><p>The Container Relocation Problem (CRP) is a key challenge in terminal operations. It aims to identify an optimal sequence for moving containers, minimizing the total number of container rehandling operations within a container yard bay. The Dynamic Container Relocation Problem (DCRP) further complicates this task this by incorporating the dynamic arrival of containers, necessitating rapid rescheduling of retrieval sequences. In this paper, we present a deep reinforcement learning framework to address both CRP and DCRP under a unified policy network. To handle dynamic layout changes, we introduce a novel dynamic attention mechanism that integrates long short-term memory and multi-head attention layers, enabling the model to adaptively capture spatial and temporal dependencies in container stacks. Trained with policy gradient optimization, our approach learns to balance immediate placement decisions with long-term retrieval efficiency without relying on domain-specific heuristics. The experiments demonstrate that our approach outperforms traditional methods for both CRP and DCRP, highlighting the potential of deep reinforcement learning in solving CRPs without extensive expert knowledge.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 16","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176374","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"pDOC-STANet: A spatiotemporal attention network for automatic assessment of prolonged disorders of consciousness using fNIRS","authors":"Zehai Li,&nbsp;Yuxin Xing,&nbsp;Shuwei Wang,&nbsp;Ke Cui,&nbsp;Xiaobo Chen,&nbsp;Huijuan Wu,&nbsp;Yuyi Feng,&nbsp;Pengfei Yi,&nbsp;Sailing He","doi":"10.1007/s10489-026-07542-1","DOIUrl":"10.1007/s10489-026-07542-1","url":null,"abstract":"<div><p>This study addresses the difficulty and subjectivity of assessing consciousness levels in patients with prolonged disorders of consciousness (pDOC) by proposing an automated classification method based on functional near-infrared spectroscopy (fNIRS). A spatiotemporal attention-based feature-fusion network, termed pDOC-STANet, was developed. The temporal branch extracts multiscale features using an improved InceptionTime architecture with squeeze-and-excitation modules, whereas the image branch represents each trial as channel-wise Gramian Angular Difference Field (GADF) maps and processes the resulting 105 × 330 × 330 tensor with a modified ResNet18. An attention-fusion module adaptively integrates the two representations. The model was evaluated using stratified, subject-independent five-fold cross-validation, with all trials from each participant kept in the same fold. pDOC-STANet achieved an accuracy of 88.72%, outperforming traditional machine-learning methods and single-branch deep-learning baselines. Ablation experiments confirmed the contributions of GADF encoding, dual-branch modeling, and attention-based fusion. LIME and SHAP analyses further indicated that the model relied on discriminative channels in bilateral frontotemporal and prefrontal regions, with HbR- and HbT-related features making substantial contributions. These results suggest that pDOC-STANet is a promising framework for fNIRS-based consciousness assessment, while external multi-center validation remains necessary before broad clinical deployment.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 16","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10489-026-07542-1.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176376","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Integrating fuzzy approximation with macrogranular information for multi-granularity feature selection","authors":"Ping Lv,&nbsp;Shaowei Yan,&nbsp;Jin Qian","doi":"10.1007/s10489-026-07540-3","DOIUrl":"10.1007/s10489-026-07540-3","url":null,"abstract":"<div><p>Feature selection, as a crucial data mining technique, is a necessary preprocessing step prior to machine learning. Fuzzy rough sets provide a solid mathematical foundation for this technique. However, most existing fuzzy rough feature selection methods rely on a single perspective—either an algebraic view that uses fuzzy dependency functions to compute feature significance, or an informational view that employs fuzzy information entropy to evaluate feature relevance. These approaches typically fail to simultaneously capture the geometric structure and uncertainty in data, resulting in limited applicability to complex datasets and suboptimal feature selection performance. To address this, this paper develops a filter-based feature selection method that integrates fuzzy approximation and macrogranular information through a multi-granularity fuzzy rough modeling framework. First, we propose an adaptive fuzzy neighborhood radius to establish multi-granularity fuzzy approximation and macrogranular entropy, enabling more refined capture of decision boundaries and information gain of features. Second, we introduce a mutual information calculation method driven by fuzzy set difference operations, which can more directly mine the relevance of features relative to the decision attribute, laying the groundwork for feature selection. Furthermore, to resolve the inconsistency in numerical scales between fuzzy approximation and macrogranular information measures, we innovatively employ an indicator function to fuse the two, proposing a multi-granularity relative significance discrimination indicator to uncover feature significance from both algebraic and informational perspectives. Finally, a multi-granularity feature selection algorithm with acceptable time complexity is designed. In the experimental section, 20 public benchmark datasets are selected and compared with 7 state-of-the-art algorithms. The results demonstrate that the proposed method achieves superior performance in improving classification accuracy. The code for this paper is publicly available at https://github.com/YANshaowei18/MGFS.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 16","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176633","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"DE-SVDD: dual-encoder SVDD for multivariate time-series anomaly detection","authors":"Xin Xie,&nbsp;Huichaoyi Zhang,&nbsp;Kexuan Liu,&nbsp;Mengqi Wu,&nbsp;Tao Wan","doi":"10.1007/s10489-026-07539-w","DOIUrl":"10.1007/s10489-026-07539-w","url":null,"abstract":"<div><p>Reconstruction-based multivariate time-series anomaly detection often relies mainly on reconstruction error, leaving the learned representation insufficiently constrained for normal-boundary modeling. We propose DE-SVDD, a dual-encoder framework that separates reconstruction-oriented representation learning from one-class boundary modeling. By imposing the SVDD constraint in a dedicated boundary space rather than directly on the reconstruction representation, DE-SVDD reduces the gradient conflict and scale imbalance between the two objectives. The framework re-encodes reconstructed sequences through a lightweight second encoder and a compact projection head, while a multi-scale context block captures temporal dependencies at different scales. Reconstruction error and center distance are then combined for anomaly scoring. Experiments on six public datasets show that DE-SVDD achieves the best F1 scores on MSL, SMAP, SMD, and PSM, reaching 0.5546, 0.5495, 0.5544, and 0.6382, respectively.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 16","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176630","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}