Large-scale deployment of AI in oncology is no longer a question of algorithmic performance alone, but of system-level safety, accountability, interoperability, and regulatory compliance. This paper proposes a reference deployment model for an integrated AI platform in a large oncology center, explicitly shaped by the constraints of the EU AI Act and the Medical Device Regulation (MDR). The paper distills and generalizes lessons learned from an almost year-long collaboration between clinical experts at NIO-PIB and engineering and research teams at UWM, conducted under a formal Letter of Intent and consolidated in a comprehensive internal project charter document for the OnkoBot program [1]. During this period, the team iteratively advanced proof-of-concept prototypes across multiple platform subsystems and formalized successive milestones via mutually agreed project charters. The proposed model is formulated to be transferable across comparable institutions while being anchored in a case-guided instantiation (OnkoBot) that illustrates practical design choices and trade-offs. It combines: (i) a multi-layer, agent-oriented reference architecture (CEMA—Community of Collaborative Evolving Medical Assistants); (ii) a trust layer (OnkoTrust) integrating measurable trust estimation, risk-aware gating, auditability, and Human-in-the-Loop escalation, conceptually grounded in Granular Computing (GrC) principles for managing admissi-ble uncertainty; (iii) knowledge grounding via RAG/GraphRAG with provenance, versioning, and update governance; and (iv) a phased deployment pathway (Preparation → Pilot → Integration → CEMA) aligned with governance and change-management practices suitable for large hospitals. The model was developed based on the authors’ combined scientific, technical, and organizational experience from delivering large-scale IT and AI initiatives in Poland and the United States, building in part on concepts and methodologies previously presented in the author’s monograph [11]. We conclude with recommended artifacts, metrics, and validation checkpoints that can support compliant, scalable, and clinically responsible AI adoption in oncology. Keywords: integrated clinical AI; reference deployment model; EU AI Act; MDR; trustworthy AI; governance; HITL; RAG; GraphRAG; oncology informatics

Granular computing is essential for data mining and knowledge discovery because real-world data is overwhelmingly complex and incomplete. Given the constraints of single-scale tables, multi-scale decision tables have attracted significant research interest. In this paper, we propose a new data analysis model called the incomplete multi-scale probabilistic covering rough set from the perspective of granular computing. This model gains some fault-tolerant characteristics by introducing a pair of probability thresholds, as well as the ability to handle uncertain and imprecise information better. In view of this, we further construct a new type of decision table, namely the incomplete multi-scale probabilistic covering decision table. For both consistent and inconsistent incomplete multi-scale probabilistic covering decision tables, we propose corresponding optimal scale selection algorithms. Furthermore, we construct an object threshold calculation approach to improve the accuracy of model employing the relative loss function. We also developed a knowledge acquisition rule algorithm based on the three-way decision concept, thereby enhancing the applicability and universality of the decision rules. Finally, the effectiveness of the proposed model and methods is verified through example analysis.

Runoff simulation based on granular computing by introducing terrain factors to construct climate characteristic index

GrC-VMM: An intelligent framework for virtual machine migration optimization using granular computing

Partial multilabel feature selection (PMLFS) is a prevalent subject that aims to enhance the performance of multilabel learning (MLL) in the context of noisy labels. In PMLFS, a crucial aspect is handling the false positive labels hidden in the candidate label set, as the imprecise annotations could mislead the feature selection process. However, many existing approaches for partial label disambiguation rely on topology information and tend to be error-prone. Besides, feature selection frameworks are often built upon a linear regression model, leading to a reliance on the classifier and a deficiency in exploring local structures. Focusing on the issues above, this article proposes a novel two-stage PMLFS method, resorting to the ideology of granular computing. In the first stage, a label disambiguation method is developed using label-specific information. Specifically, a specific granular ball computing model is designed to characterize the distribution of datapoints labeled differently, and therefore, using the affinity relationships among samples and balls, the label-specific information concealed in the data distribution can be captured for label disambiguation. In the second stage, a filter-based feature selection method that explores the local structure of samples is presented. This method relies on a devised fuzzy decision neighborhood rough set (FDNRS) to capture more detailed membership information by maximizing the neighborhood consistency of samples' related labels. Simultaneously, the feature selection method minimizes the uncertainty derived from unrelated labels. Extensive experiments on 12 datasets in terms of four evaluation metrics demonstrated the effectiveness of the proposed approach.

Research on modelling geographical accessibility to healthcare services has witnessed rapid methodological advancement and refinement. One of the contributing factors is the increasing availability of big data detailing the link between the population in need of care and the health facility such as infrastructure, travel modes and speeds, traffic congestion and the quality of road network. This has allowed more granular computation of geographic access metrics, particularly in low-and-middle income countries where data are scarce. However, there are no reviews providing a comprehensive overview of the availability and use of big data for assessing geographical accessibility to healthcare. This protocol aims to describe a methodological approach that will be used to review the existing literature on the application of big data (past or potential) in evaluating geographical accessibility to healthcare. To characterise the big data that can be used to model geographical accessibility to healthcare, a scoping review will be undertaken and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extensions for Scoping Reviews guidelines. We will search seven scientific databases (PubMed, Scopus, Web of Science, EBSCOhost-CINAHL, Cochrane, Embase and MEDLINE via Ovid), grey literature, reference lists of identified publications and conference proceedings. Search engines will be used to identify relevant big data services not yet used in published academic literature. All literature published in English or French will be included, regardless of publication type, geographical location or year of publication provided it describes or mentions big data that may be useful for evaluating geographical accessibility to healthcare. Study selection and data extraction will be performed independently by two researchers with a third resolving any discrepancies. Analysis will be conducted to summarise big data providers, their characteristics and their usefulness in terms of types of spatial accessibility metrics that can be derived. Formal ethical approval is not required, as primary data will not be collected in this review. Findings will be disseminated through peer-reviewed publication in a journal, conference presentation and condensed summaries for stakeholders through professional networks and social media summaries. Open Science Framework (OSF): https://doi.org/10.17605/OSF.IO/S496F.

In this study, a Granular Function Fractional-Order Proportional-Integral-Derivative (GFFOPID) controller is developed for nonlinear suspension systems. Initially, a Fractional Order Proportional-Integral-Derivative (FOPID) controller is employed, and its parameters are tuned using fuzzy systems to enhance performance under disturbances and uncertainties. Subsequently, Particle Swarm Optimization is applied to further optimize these parameters for improved control efficacy. The Fractional Order fuzzy Proportional-Integral-Derivative controller, which involves fuzzy sets and inference process, can suffer from extended computation times. This issue is mitigated through the introduction of granular computing, which simplifies the fuzzy control process by substituting traditional inference with granular sampling functions. Experimental results show that the GFFOPID controller effectively improves both the maneuvering stability and ride comfort of the vehicle. Furthermore, the use of granular functions effectively resolves the rule base explosion issue, reduces computational complexity, and significantly enhances the controller’s efficiency.

ABSTRACT Accurate river flow prediction is critical for sustainable water resource management, particularly in arid and semi-arid regions. However, balancing model accuracy, computational efficiency, and interpretability remains a significant challenge due to the complex and nonlinear nature of hydrological systems. This study employs a granular computing (GRC) model to predict monthly inflows to the Alavian Dam in Iran. Principal component analysis (PCA) was used to reduce input dimensionality, identifying six key variables to enhance computational performance. The predictive performance of GRC was compared with artificial neural networks (ANNs) and support vector machines (SVMs), using R2, RMSE, and MAE as evaluation metrics. The GRC model achieved R2 values of 0.93 during calibration and 0.94 during validation, outperforming both ANN and SVM. Notably, GRC demonstrated superior accuracy in capturing extreme flow events, which are crucial for flood and drought management. This advantage is attributed to its rule-based structure and local learning approach, which enables effective modeling of nonlinearities and sparse data. Furthermore, the interpretability of the GRC model - facilitated by its use of granules and transparent if-then rules - offers valuable insights into variable influence. These strengths highlight GRC as a reliable and efficient tool for hydrological forecasting and climate-adaptive water resource planning.

Abstract The goal of outlier detection is to pinpoint data points that exhibit notable deviations from the rest of the observed values. It has found successful application in numerous fields, including process inspection, anti-terrorist operations, and public security. However, the majority of existing outlier algorithms rely on methods involving filling or deleting missing data, with few directly addressing incomplete data. This article studies outlier detection for incomplete real-valued data based on class-consistent technology, rough set theory, and granular computing. First, a tolerance relation founded on class-consistent technology is presented to illustrate the similarity among information values within an incomplete real-valued information system (IRVIS). Then, the tolerance classes are established based on the tolerance relation and utilized for computing approximate accuracy and other metrics. Next, an outlier factor is defined, considering both the degree of outlierness and the weight function assigned to each object within an IRVIS, elucidating the uncertainty and degree of outlier. Finally, an outlier detection algorithm (ODIRG) for an IRVIS based on class-consistent technology, rough set theory, and granular computing is devised. Numerical experiments on seven UCI datasets are undertaken to evaluate the stability of the ODIRG algorithm. The proposed method is demonstrated to exhibit strong effectiveness and adaptability for categorical data when compared with five other algorithms. It is notable that for comprehensive comparison, precision, recall, F F 1-measure, and receiver operating characteristic curve are employed to delineate the benefits of the proposed approach.

Maintaining optimal microclimatic conditions within greenhouses represents a significant challenge in modern agricultural contexts, where prediction systems play a crucial role in regulating temperature and humidity, thereby enabling timely interventions to prevent plant diseases or adverse growth conditions. In this work, we propose a novel approach which integrates a cascaded Feed-Forward Neural Network (FFNN) with the Granular Computing paradigm to achieve accurate microclimate forecasting and reduced computational complexity. The experimental results demonstrate that the accuracy of our approach is the same as that of the FFNN-based approach but the complexity is reduced, making this solution particularly well suited for deployment on edge devices with limited computational capabilities. Our innovative approach has been validated using a real-world dataset collected from four greenhouses and integrated into a distributed network architecture. This setup supports the execution of predictive models both on sensors deployed within the greenhouse and at the network edge, where more computationally intensive models can be utilized to enhance decision-making accuracy.

This study is about what matters: predicting when microfinance institutions might fail, especially in places where financial stability is closely linked to economic inclusion. The challenge? Creating something practical and usable. The Adjusted Gross Granular Model (ARGM) model comes here. It combines clever techniques, such as granular computing and machine learning, to handle messy and imbalanced data, ensuring that the model is not just a theoretical concept but a practical tool that can be used in the real world.Data from 56 financial institutions in Peru was analyzed over almost a decade (2014-2023). The results were quite promising. The model detected risks with nearly 90% accuracy in detecting failures and was right more than 95% of the time in identifying safe institutions. But what does this mean in practice? It was tested and flagged six institutions (20% of the total) as high risk. This tool's impact on emerging markets would be very significant. Financial regulators could act in advance with this model, potentially preventing financial disasters. This is not just a theoretical exercise but a practical solution to a pressing problem in these markets, where every failure has domino effects on small businesses and clients in local communities, who may see their life savings affected and lost due to the failure of these institutions. Ultimately, this research is not just about a machine learning model or using statistics to evaluate results. It is about giving regulators and supervisors of financial institutions a tool they can rely on to help them take action before it is too late when microfinance institutions get into bad financial shape and to make immediate decisions in the event of a possible collapse.

In recent years, with the rapid development of the Internet, multisource information fusion has become a forefront issue due to its ability to merge different information. Granular computing (GrC), as a methodology simulating human hierarchical cognition, provides a new approach for multisource information fusion. However, on one hand, the existing information fusion studies in GrC all focus on feature-level fusion and decision-level fusion based on multisource data, neglecting the basic characteristics and advantages of GrC: granulation. On the other hand, the existing methods for fusing the knowledge spaces in GrC suffer from losing the necessary information or artificially adding information. In order to address these issues, a novel information fusion mode from the perspective of GrC is proposed in this article, named knowledge-level fusion. First, by introducing a new step, that is, granulate data to construct the knowledge space, into the multisource information fusion process, the knowledge-level fusion mode is proposed. Second, the optimistic core quotient space is proposed to characterize the information consensus and information gap of multisource knowledge spaces in the static data environment. The pessimistic core quotient space is proposed to characterize the information consensus in the dynamic data environment. Related theorems are given to describe the characteristics of the core quotient spaces. Then, the knowledge-level fusion method driven jointly by the data space and the knowledge space is introduced based on the principle of extracting the core quotient space first and then allocating other objects in the candidate set. On the basis, the superiority of the proposed method over the existing methods is demonstrated through theoretical analysis. Finally, experiments on 12 UCI datasets and three UKB datasets are carried out to verify the promoting effect on classification and clustering algorithms, the effectiveness compared to feature-level and decision-level fusion modes, efficiency and statistical significance of the proposed knowledge-level fusion method and mode.

In the era of Big data, how to filter massive information and push it to appropriate users is a subject that has been explored in computer information technology. In this context, in view of the poor effect of educational game recommendation affected by data sparsity, a collaborative filtering recommendation (CFR) algorithm integrating the covering rough granular layer clustering (CRGLC) and K-means clustering is proposed. On the basis of the K-means clustering CFR model, granular computing (GC) is introduced to build a covering rough granular space (CRGS) based on the user's comprehensive score and game type. By setting and adjusting coverage coefficients, local rough particle (LRP) sets of game users are searched under different particle layers to mitigate the impact of data sparsity. The improved algorithm is tested on the data sets with the sparsity of 0.937 and 0.901, and the mean absolute error (MAE) values of the two are 0.708 and 0.716. The results are relatively close, indicating that the model can effectively improve the accuracy of the model in the case of sparse data. Research is organized on the classification accuracy of the model, and the accuracy and F1 scores are 0.880 and 0.826, which are higher than the social spatial-temporal probabilistic matrix factorization and Slope One models in the literature. This indicates that the model is more accurate in identifying and classifying game types and is more conducive to educational game recommendations. In practical application performance testing, the model has small and large intra-cluster variations, resulting in good clustering performance. Compared with the known and better recommended algorithms of attributes clustering and score matrix filtering, dynamic evolutionary collaborative filtering, double trace normal minimization, and evolutionary heterogeneous clustering collaborative filtering, its MAE and root-mean-square error of scoring prediction are the lowest. By using this model to predict ratings for 500 user samples, the error is only 2.8%. It has been proved that this algorithm has higher accuracy in educational game recommendations. Overall, the innovation of the algorithm lies in the fusion of CRGLC and K-means clustering, and the introduction of grain computing to deal with the data sparsity problem and improve the recommendation accuracy. This research has some practical value in solving the problem of sparse data in educational game recommendations.

Class incremental learning (CIL) is a specific scenario in incremental learning. It aims to continuously learn new classes from the data stream, which suffers from the challenge of catastrophic forgetting. Inspired by the human hippocampus, the CIL method for replaying episodic memory offers a promising solution. However, the limited buffer budget restricts the number of old class samples that can be stored, resulting in an imbalance between new and old class samples during each incremental learning stage. This imbalance adversely affects the mitigation of catastrophic forgetting. Therefore, we propose a novel CIL method based on multi-granularity balance strategy (MGBCIL), which is inspired by the three-way granular computing in human problem-solving. In order to mitigate the adverse effects of imbalances on catastrophic forgetting at fine-, medium-, and coarse-grained levels during training, MGBCIL introduces specific strategies across the batch, task, and decision stages. Specifically, a weighted cross-entropy loss function with a smoothing factor is proposed for batch processing. In the process of task updating and classification decision, contrastive learning with different anchor point settings is employed to promote local and global separation between new and old classes. Additionally, the knowledge distillation technology is used to preserve knowledge of the old classes. Experimental evaluations on CIFAR-10 and CIFAR-100 datasets show that MGBCIL outperforms other methods in most incremental settings. Specifically, when storing 3 exemplars on CIFAR-10 with Base2 Inc2 setting, the average accuracy is improved by up to 9.59% and the forgetting rate is reduced by up to 25.45%.

Granule description is a fundamental problem in granular computing. However, how to describe indefinable granules is still an open, interesting, and important problem. The main objective of this paper is to give a preliminary solution to this problem. Before proceeding, the framework of approximate description is introduced. That is, any indefinable granule is characterized by an ordered pair of formulas, which form an interval set, where the first formula is the β-prior approximate optimal description and the second formula is the α-prior approximate optimal description. More concretely, given an indefinable granule, by exploring the description of its lower approximate granule, its β-prior approximate optimal description is obtained. Likewise, by consulting the description of its upper approximate granule, its α-prior approximate optimal description can also be derived. Following this idea, the descriptions of indefinable granules are investigated. Firstly, ∧-approximate descriptions of indefinable granules are investigated based on the classical concept lattice, and (∧,∨)-approximate descriptions of indefinable granules are given via object pictorial diagrams. And then, it is revealed from some examples that the classical concept lattice is no longer effective and negative attributes must be taken into consideration. Therefore, a three-way concept lattice is adopted instead of the classical concept lattice to study (∧,¬)-approximate descriptions and (∧,∨,¬)-approximate descriptions of indefinable granules. Finally, some discussions are presented to show the differences and similarities between our study and existing ones.

Action Rules are rule based systems that extract actionable patterns which are hidden in big volumes of data. Huge amount of data gets generated from Education sector, Business field, Medical domain and Social Media, in a single day. In the technological world of big data, massive amounts of data are collected by organizations, including in major domains like financial, medical, social media and Internet of Things(IoT). Mining this data can provide a lot of meaningful insights on how to improve user experience in multiple domain. Users need recommendations on actions they can undertake to increase their profit or accomplish their goals, this recommendations are provided by Actionable patterns. For example: How to improve student learning; how to increase business profitability; how to improve user experience in social media; and how to heal patients and assist hospital administrators. Action Rules provide actionable suggestions on how to change the state of an object from an existing state to a desired state for the benefit of the user. The traditional Action Rules extraction models, which analyze the data in a non distributed fashion, does not perform well when dealing larger datasets. In this work we are concentrating on the vertical data splitting strategy using information granules and creating the data partitioning more logically instead of splitting the data randomly and also generating meta actions after the vertical split. Information granules form basic entities in the world of Granular Computing(GrC), which represents meaningful smaller units derived from a larger complex information system. We introduced Modified Hybrid Action rule method with Partition Threshold Rho. Modified Hybrid Action rule mining approach combines both these frameworks and generates complete set of Action Rules, which further improves the computational performance with large datasets.

Action Rules are rule based systems that extract actionable patterns which are hidden in big volumes of data. Huge amount of data gets generated from Education sector, Business field, Medical domain and Social Media, in a single day. In the technological world of big data, massive amounts of data are collected by organizations, including in major domains like financial, medical, social media and Internet of Things(IoT). Mining this data can provide a lot of meaningful insights on how to improve user experience in multiple domain. Users need recommendations on actions they can undertake to increase their profit or accomplish their goals, this recommendations are provided by Actionable patterns. For example: How to improve student learning; how to increase business profitability; how to improve user experience in social media; and how to heal patients and assist hospital administrators. Action Rules provide actionable suggestions on how to change the state of an object from an existing state to a desired state for the benefit of the user. The traditional Action Rules extraction models, which analyze the data in a non distributed fashion, does not perform well when dealing larger datasets. In this work we are concentrating on the vertical data splitting strategy using information granules and creating the data partitioning more logically instead of splitting the data randomly and also generating meta actions after the vertical split. Information granules form basic entities in the world of Granular Computing(GrC), which represents meaningful smaller units derived from a larger complex information system. We introduced Modified Hybrid Action rule method with Partition Threshold Rho. Modified Hybrid Action rule mining approach combines both these frameworks and generates complete set of Action Rules, which further improves the computational performance with large datasets.

Undoubtedly, machine learning (ML) has demonstrated a wealth of far-reaching successes present both at the level of fundamental developments, design methodologies and numerous application areas, quite often encountered in domains requiring a high level of autonomous behavior. Over the passage of time, there are growing challenges of privacy and security, interpretability, explainability, confidence (credibility), and computational sustainability, among others. In this study, we advocate that these quests could be addressed by casting them both conceptually and algorithmically in the unified environment augmented by the principles of granular computing. It is demonstrated that the level of abstraction, delivered by granular computing plays a pivotal role in the interpretation by quantifying the level of credibility of ML constructs. The study also highlights the principles of granular computing and elaborates on its landscape. The original idea of a comprehensive and unified framework of data-knowledge environment of ML is introduced along with a detailed discussion on how data and knowledge are used in a seamless fashion by invoking granular embedding and producing relevant loss functions. Key categories of knowledge-data integration realized at the levels of data and model (involving symbolic/qualitative models and physics-oriented models) and investigated.

Minimum Cost Consensus Model Based on Granular Computing Through Online Reviews for Supporting Group Decision-Making

Distance-based granular computing in networks modeled by intersection graphs