The rapid growth of blockchain technology and the increasing number of network nodes have heightened the risk of sophisticated attacks. Among these, Eclipse attacks present a serious threat to decentralized networks by exploiting their peer-to-peer structures. While previous research has explored artificial intelligence techniques to defend against Eclipse attacks, evolving attack patterns continue to challenge existing defenses. In this paper, we propose a novel defense framework that integrates a clustering approach based on self-attention encoders within a multi-kernel neural network clustering model. Our method utilizes parallel subnetworks to extract category-specific features from multiple perspectives, generating discriminative cluster centroids that are combined with raw transaction data to train a robust classifier for detecting Eclipse attacks in Ethereum networks. To evaluate our approach, we simulate Eclipse attacks on the Ethereum testnet and conduct extensive experiments. The results demonstrate that our method achieves a detection accuracy of 98.5% and improves classification performance by 5% compared to models trained without cluster-enhanced features, confirming the effectiveness of the proposed defense.

A density-driven graph-based clustering model with adaptive outlier recognition

Elucidation of molecular interactions between selected phenylpropanoids and myofibrillar protein: Conformational remodeling and machine learning analysis.

HierFLMC: Efficient hierarchical federated learning based on soft clustering model compression

BackgroundIntegrating brain structure and function may help characterize neurobiological heterogeneity in migraine alongside symptom presentation.AimTo apply a multimodal, exploratory, data-driven approach to identify migraine subgroups using structural and functional MRI, and to describe the clinical characteristics of the resulting subgroups.MethodsResting-state functional connectivity (FC) across cortical and subcortical regions, along with structural measures including cortical thickness, cortical volume, and subcortical volumes, were extracted from 111 individuals with migraine (75 chronic, 36 episodic) classified according to ICHD-3 criteria. After dimensionality reduction using principal component analysis, hierarchical agglomerative clustering was applied to identify multimodal imaging-derived subgroups. For comparison, secondary unimodal clustering models were constructed using functional-only and structural-only feature sets. The optimal number of clusters was determined using silhouette coefficients, and clustering concordance across models was quantified using the Adjusted Rand Index (ARI). Group differences in clinical characteristics, FC, and cortical and subcortical structure were assessed using covariate-adjusted statistical models with false discovery rate (FDR) correction.ResultsMultimodal clustering identified two subgroups with distinct clinical and imaging profiles, Migraine Cluster 1 (M1f + s) and Migraine Cluster 2 (M2f + s). M2f + s showed older age, longer disease duration, greater migraine disability, widespread increases in cortical-subcortical FC (including Dorsal Attention, Somatomotor, and Visual networks), and reduced cortical volumes across frontal, parietal, temporal, and insular regions compared with M1f + s. This subgroup also exhibited increased connectivity relative to controls. In contrast, M1f + s showed preserved cortical structure and stronger Control-network-subcortical connectivity compared to M2f + s, and no significant functional or structural deviations from controls. Unimodal analyses revealed that Functional-only clustering aligned moderately with the multimodal cluster solution (ARI = 0.427), showing that FC was a primary determinant of the multimodal cluster structure, whereas structural-only clustering showed negligible overlap (ARI = 0.001), reflecting an orthogonal dimension of heterogeneity captured by structural variation.ConclusionData-driven multimodal neuroimaging-based clustering in migraine identified two subgroups with distinct clinical and imaging patterns, highlighting heterogeneity and providing a framework for further investigation of imaging-informed characterization.

Development of a Regression-Based Overlapping Clustering Model for Assessing Harbor Area Adequacy in National Fishing Ports

To address the challenges of traditional methods for monitoring rock dynamic hazards in mines, which struggle to fully characterize the spatiotemporal heterogeneity of damage evolution and the resulting lag in early warning, this paper proposes a dynamic rock damage classification and fracture early warning model driven by acoustic emission data. Based on an improved dynamic K-means algorithm, this model fuses time dependence, energy intensity, and event spatial density characteristics through exponentially decaying weights to construct a spatiotemporal-energy synergistic clustering framework. Furthermore, a nonlinear coupling model for the comprehensive risk index (RI) is established, combining the static damage variable D with dynamic parameters such as energy release rate, ring count, and spatial clustering, to create a five-level early warning threshold. Experimental results demonstrate that the improved algorithm achieves clustering silhouette coefficients exceeding 0.7 for single-source, multi-source, and complex fracture patterns, and the error between cluster regions and actual fracture distribution is less than 1 mm. The RI model accurately identifies the damage state of the test block and effectively predicts critical instability, significantly improving both timeliness and accuracy. This research overcomes the limitations of traditional static evaluation and provides high-precision technical support for real-time monitoring of hidden rock fractures and prevention and control of mine dynamic hazards.

Groundwater modeling in semi-arid regions presents significant challenges due to complex aquifer dynamics, limited data availability, and heterogeneous hydrogeological conditions. This study presents a comprehensive comparative analysis of three fuzzy expert system strategies for monthly groundwater level forecasting in the Al-Hsain Basin, Syria: localized models based on hydrogeographical grouping, a unified basin-wide approach, and an innovative behavioral clustering methodology. Using synchronized rainfall and temperature data from 35 monitoring wells over four years (2020–2024), we developed and evaluated fuzzy inference systems’ directional classification accuracy as the primary performance metric, categorizing groundwater level changes into rise, stable, and decline states rather than predicting continuous values. This choice reflects the qualitative nature of fuzzy expert systems and their suitability for groundwater management under data-limited conditions. The behavioral clustering approach achieved excellent overall performance with a mean accuracy of 0.74, outperforming localized models (0.71) and unified models (0.67). Behavioral clustering demonstrated effectiveness in 66% of wells, with individual accuracy improvements reaching up to 0.23, while reducing model complexity from five group-specific systems to three behaviorally coherent clusters. Localized models achieved optimal performance in 29% of wells where hydrogeological conditions aligned with spatial assumptions, whereas unified models provided consistent moderate performance across 89% of locations. The incorporation of lagged variables and seasonal indices in behavioral clustering models proved essential for capturing temporal complexity in semi-arid groundwater responses. Statistical analysis revealed lower intra-group variability in behavioral clusters (standard deviation 0.06–0.09) than in geographical groupings (0.08–0.14), confirming improved functional homogeneity through response-based organization. These findings indicate that fuzzy modeling strategy selection should be context-dependent, with behavioral clustering offering an effective balance between accuracy, interpretability, and generalization for regional groundwater management applications. The novelty of this work lies in isolating the effect of fuzzy system organization logic (localized, unified, and behavioral) on forecasting performance, robustness, and transferability, evaluated under an identical inference and time-series validation framework.

The DISC framework is widely used to describe behavioral styles in organizations, but it is often applied through static and qualitative interpretation. This study combines graph-based clustering with supervised learning to analyze DISC-style profiles, competencies, and stress outcomes. Using a real-world dataset of 195 employees described by 97 heterogeneous attributes, we construct a weighted similarity graph by fusing (i) cosine similarity of 17 ordinal competency levels, (ii) exact-match similarity of organizational context variables, and (iii) Jaccard similarity of trait-like descriptors. Modularity-based community detection is applied to reveal latent behavioral groups. Random Forest models are then used to predict stress-related outcomes. For 4-class stress prediction (Low, Medium, High, High (Work-related)), stratified 5-fold cross-validation yields an average accuracy of 52.82%. This is above the uniform random baseline (25%) but below the majority-class baseline ($$\approx 58.97\%$$), indicating moderate predictive signal. Variable-importance analysis suggests that sales-related competency levels contribute strongly to stress differentiation in this cohort. A separate experiment on competency-group prediction reaches near-perfect accuracy, but this is expected because the target is derived from the same competency descriptors used as inputs and therefore reflects information leakage rather than generalizable prediction. Overall, the study shows how DISC assessments can be extended into graph-based and predictive organizational analytics, while also clarifying the limits of what can be inferred from cross-sectional survey attributes.

With the continuous expansion of user scale on social media platforms, WeChat Moments, as a core social scene for Chinese users, has become a key issue in privacy management and user experience optimization due to the dynamic monitoring of interface blocking behavior and the analysis of generational differences. Current research faces three limitations: Firstly, traditional object detection techniques struggle to meet the real-time requirements of dynamic interface operations, especially in the detection of small targets, resulting in significant accuracy losses. Secondly, existing behavior analysis methods rely on log statistics and questionnaire surveys, lacking the ability to model multimodal interactive behavior in both time and space. Thirdly, user portrait construction is mostly based on static attribute classification, failing to effectively capture the dynamic differences in privacy strategy selection among generational groups. To address these challenges, this study proposes a multimodal collaborative analysis framework that integrates an improved YOLOv5 architecture with dynamic portrait generation. At the methodological level, a three-level collaborative computing architecture is designed. A lightweight YOLOv5-GhostNet model is deployed on mobile devices, achieving cross-modal feature decoupling of text, image, and video blocking behavior through a multi-scale dilated convolution pyramid and dynamic weight fusion mechanism. The detection accuracy reaches 93.2%, an improvement of 8.1% over the baseline model. Secondly, a dynamic threshold algorithm with composite elastic windows is proposed, combining event density perception and dual attenuation factors to reduce the false trigger rate to 4.2%, while simultaneously optimizing the real-time response delay to 105 ms. Furthermore, an orthogonal constrained multimodal fusion strategy is introduced, utilizing KL divergence feature selection and an XGBoost clustering model to construct generational sensitive behavior fingerprints. This reveals behavioral patterns such as Generation Z users preferring fine-grained permission control (58.3% partial visibility) and low complexity in operation paths (single session duration of 1.5 s), forming a significant differentiation from Generation X users’ dominant strategy of complete blocking (68.7%). Experiments show that the system maintains a false trigger rate of only 5.1% under adversarial attack scenarios and maintains a clustering accuracy of 83.4% even with 30% data loss. The research conclusion points out that a technical path based on dynamic feature optimization and spatiotemporal correlation modeling can effectively break through the real-time bottleneck of social interface behavior analysis. The generation of generational portraits requires the integration of cross-modal semantic decoupling and incremental learning mechanisms to cope with the coupled effects of user cognitive evolution and system iteration. This achievement provides theoretical support and technical paradigms for the optimization of privacy management strategies on social platforms. Its lightweight architecture design (model parameter count of 1.2 M) and multimodal decoupling method have universal reference value for intelligent human–computer interaction systems.

ABSTRACT Most research on the spatial organization of employment focuses on individual metropolitan areas, limiting insight into cross‐regional and generalizable patterns. At the same time, recent studies often rely on broad indicators of employment dispersion—such as the share of urbanized land occupied by jobs—while methodological advances in identifying job centers have lagged. As a result, comparative analyses of regional employment geographies, particularly those focused on job center formation and dominance, remain scarce. Using block‐level employment data, we adapt a spatial clustering model to identify job centers across the 100 most populous regions of the United States at three relative density thresholds. We then document variation in employment centralization and concentration across regions and density strata, and test—via regression models—how these patterns relate to metropolitan age, industry composition, and political fragmentation, serving as proxies for car orientation, agglomeration dynamics, and inter‐municipal competition, respectively. Our findings show that professional services are strongly associated with dense employment clustering, health care tends to agglomerate at lower densities, and the regional prevalence of finance is linked to greater employment concentration and central business district (CBD) dominance. Additionally, denser regions tend to have fewer but more dominant job centers, and political fragmentation is associated with flatter job center hierarchies and weaker CBD primacy.

Structural health monitoring (SHM) of civil structures, especially bridges, using vibration data is a practical approach for ensuring their functionality and integrity. However, long-term vibration monitoring of these structures faces challenges due to seasonal environmental variations. This study proposes an innovative machine learning method to enhance SHM of bridges subjected to environmental variability. Developed from the concept of unsupervised learning, this method combines teacher–student clustering with regional anomaly detection. The proposed dual-model framework features a sophisticated parametric clustering model (i.e., the teacher) that guides a simpler non-parametric clustering model (i.e., the student) to generate localized data. This setup improves data segmentation and produces localized training subsets for anomaly detector modeling. These local data are then used to train a regional one-class support vector machine, which serves as the main anomaly detector for SHM. This model computes anomaly indices by effectively isolating genuine structural anomalies from patterns of environmental variability. The major contributions of this research are twofold: it introduces an unsupervised learning solution for long-term SHM amid significant environmental changes in vibration features and integrates clustering and anomaly detection to develop a new hybrid framework, thereby enhancing the reliability of monitoring programs. Long-term modal frequencies of large-scale bridge structures are used to validate the proposed method, supported by several comparative analyses. Results indicate that the proposed method not only mitigates environmental variability but also provides reliable decision-making.

The rapid growth of cryptocurrency markets has created new challenges in understanding and predicting the structural dynamics of digital asset prices. Bitcoin, as the most traded blockchain-based currency, exhibits extreme volatility, nonlinear patterns, and complex regime shifts that traditional financial models cannot adequately capture. This study proposes a hybrid analytical framework that integrates K Means clustering with the Hidden Markov Model to identify and model multiple market regimes in Bitcoin time series data. The Bitcoin dataset used in this research contains minute-level records that were preprocessed to extract key indicators, namely logarithmic returns and rolling volatility, which represent the short-term dynamics of market behavior. The K Means algorithm was first employed to segment the data into three distinct clusters that correspond to bullish, bearish, and sideways regimes, followed by the application of the Hidden Markov Model to estimate probabilistic transitions between these regimes over time. The results reveal that the hybrid K Means and Hidden Markov Model approach achieves superior performance compared to a standalone model, as indicated by a higher log likelihood and a lower Bayesian Information Criterion value. The transition probability matrix shows that bullish and bearish regimes are highly persistent, while the sideways regime acts as a transitional buffer that connects both market extremes. The empirical findings confirm that Bitcoin prices evolve through persistent and probabilistically determined regimes rather than random fluctuations. The proposed framework provides a more comprehensive understanding of cryptocurrency market dynamics and offers practical value for investors, risk analysts, and policymakers in designing adaptive trading and risk management strategies within blockchain-based financial ecosystems.

ABSTRACT Accurate and explainable modelling of Autonomous Complex Systems (ACS) is vital for civil and military applications, requiring a balance between model performance and interpretability. This study proposes a nested interior–exterior optimisation approach for explainable ACS modelling, leveraging a cross‐clustered framework with migration and cooperative learning. For interior optimisation of the ACS model, a cross‐clustered migration mechanism is designed to incorporate pseudogenes to simultaneously optimise the structure and parameters, enabling compact yet accurate ACS models. For exterior optimisation of the ACS model, a cross‐clustered cooperative learning mechanism is designed to allocate additional resources to superior‐performing models and redistribute their optimal structures and parameters across different models in varied clusters. The belief rule base (BRB) is adopted as the baseline model, ensuring explainability and accessibility for decision‐makers. A case study on drone system modelling validates the approach's efficacy, demonstrating that: (1) cross‐clustered migration facilitates efficient simultaneous optimisation, producing compact ACS models with superior performance, that is, BRBs with seven and eight rules are identified as optimal ACS models for modelling two objectives; (2) cross‐clustered cooperative learning leverages multiple optimisation algorithms (e.g., GrEA, MOEA/D, SPEA2) to outperform single‐algorithm approaches and (3) comprehensive analysis of interior and exterior optimisation confirms the framework's robustness. Last but not least, optimal BRBs are fully explainability to ACS decision‐makers as it can provide completely transparent and accessible modelling procedures and inferencing interfaces.

Efficient software defect prediction using fuzzy K-member clustering and metaheuristic-driven ensemble feature learning model

This study introduces a novel classification indicator, the Feature-Weighting Speed (FWS), and a hybrid clustering approach integrating DBSCAN and Natural Breaks to address the challenge of spatial heterogeneity in traffic state classification. Key findings indicate that FWS significantly reduces cross-city variability in data distribution, as evidenced by a low Coefficient of Variation, thereby ensuring consistent classification performance across diverse urban environments. The hybrid clustering model achieves precise traffic state separation, with DBSCAN identifying thresholds for congested and free-flow speeds and Natural Breaks delineating boundaries for intermediate states. Empirical validation using large-scale data from five Chinese cities confirms that the proposed method not only achieves Davies–Bouldin Index values below 0.5 across all datasets, outperforming traditional clustering algorithms, but also demonstrates strong adaptability to varying traffic conditions. These findings underscore the practical value of the proposed framework for real-world traffic management, enabling accurate congestion pattern identification and supporting data-informed decision-making.

Background/Objectives: To identify distinct sonographic phenotypes of complex malformations of the fetal ventral wall. Methods: We performed a retrospective analysis of ultrasound reports from 160 fetuses diagnosed with complex ventral wall defects at a single tertiary referral center between 1997 and 2021. Agglomerative hierarchical clustering was applied to identify distinct sonographic phenotypes based on the level of the ventral wall defect and associated anomalies. Results: Ventral wall defects involved the abdominal wall in 150 cases, the thoracic wall in 42 cases, and the pelvic wall in 28 cases, either in isolation or in combination. Open neural tube defects were present in 58 fetuses (36.3%), spinal defects in 110 fetuses (68.8%), and limb anomalies in 45 fetuses (28.1%). Additional anomalies were identified in 38 fetuses (23.8%), including cardiac anomalies in 18 cases (11.3%). Amniotic bands were observed in seven cases (4.4%). Using agglomerative hierarchical clustering, five groups of fetuses with differing numbers of observations were identified (cluster 1, n = 104; cluster 2, n = 5; cluster 3, n = 30; cluster 4, n = 10; cluster 5, n = 11). The silhouette score of the clustering model was 0.3285. The most discriminative features for each cluster, expressed as feature importance values, were as follows: kyphoscoliosis for cluster 1 (0.924), pelvic wall defect for cluster 2 (0.852), ectopia cordis for cluster 3 (0.662), limb anomalies for cluster 4 (0.767), and spina bifida for cluster 5 (0.691). Conclusions: Complex malformations of the fetal ventral wall are associated with a wide spectrum of additional anomalies. Hierarchical clustering identified five distinct sonographic phenotypes of complex ventral wall defects, highlighting the heterogeneity of these conditions.

ABSTRACT Rowing technique to achieve optimal boat velocity depends on individual rowing style. Traditionally, quantification of rowing technique has involved discrete point analysis, limiting the understanding and interpretation of the stroke cycle with data loss occurring between the reported metrics. However, higher dimensional statistical approaches, such as functional data analysis (FDA), can facilitate enhanced understandings of temporal patterns within time series data such as force and acceleration profiles. The aim of the study was to distinguish technique characteristics during single sculling using a novel functional clustering method considering the whole stroke cycle for analysis. Twenty-five elite rowers (12 females, 25 ± 2.5 years and 13 males, 27 ± 2.8 years) completed an on-water single sculling biomechanics assessment. Gate force, foot-stretcher force and boat acceleration were independently fitted with a clustering model, with separate models created for each gender. Boat acceleration exhibited the most variability of the three independent variables in cluster group patterns and individual rowers. Results revealed more than one approach to achieving optimal boat velocity at the elite level and technical coaching strategies should be based on the individual rather than attempting to replicate successful elite rowers’ technique who may exhibit a different set of physical and technical attributes.

Morbid obesity is a complex, multifactorial disorder characterized by metabolic and inflammatory dysregulation. The aim of this study was to observe changes in obese patients adhering to a personalized nutrition plan based on multi-omic data. This study included 14 adult patients with a body mass index (BMI) > 40 kg/m2 who were consecutively recruited from those presenting to our outpatient clinic and who met the inclusion criteria. Clinical, biochemical, hormonal, and glycomic parameters were assessed, along with whole-genome sequencing (WGS) that included a focused analysis of obesity-associated genes and an extended analysis encompassing genes related to cardiometabolic disorders, hereditary cancer risk, and nutrigenetic profiles. Patients were stratified into nutrigenetic clusters using a patented unsupervised machine learning platform (German Patent Office, No. DE 20 2025 101 197 U1), which was employed to generate personalized nutrigenetic dietary recommendations for patients with morbid obesity to follow over a six-month period. At baseline, participants exhibited elevated glucose, insulin, homeostatic model assessment for insulin resistance (HOMA-IR), triglycerides, and C-reactive protein (CRP) levels, consistent with insulin resistance and chronic low-grade inflammation. The majority of participants harbored risk alleles within the fat mass and obesity-associated gene (FTO) and the interleukin-6 gene (IL-6), together with multiple additional significant variants identified across more than 40 genes implicated in metabolic regulation and nutritional status. Using an AI-driven clustering model, these genetic polymorphisms delineated a uniform cluster of patients with morbid obesity. The mean GlycanAge index (56 ± 12.45 years) substantially exceeded chronological age (32 ± 9.62 years), indicating accelerated biological aging. Following a six-month personalized nutrigenetic dietary intervention, significant reductions were observed in both BMI (from 52.09 ± 7.41 to 34.6 ± 9.06 kg/m2, p < 0.01) and GlycanAge index (from 56 ± 12.45 to 48 ± 14.83 years, p < 0.01). Morbid obesity is characterized by a pro-inflammatory and metabolically adverse molecular signature reflected in accelerated glycomic aging. Personalized nutrigenetic dietary interventions, derived from AI-driven analysis of whole-genome sequencing (WGS) data, effectively reduced both BMI and biological age markers, supporting integrative multi-omics and machine learning approaches as promising tools in precision-based obesity management.

Weaning is a critical stage in swine production, characterized by intestinal alterations that affect piglet health and performance. In this study, machine learning techniques were applied to identify joint patterns between gut health and productivity during the first 15 days post-weaning. A total of 103 animals were analyzed using a dataset of 24 histomorphological, biochemical, and productive variables. Among the unsupervised clustering models, K-means (k = 2) achieved the best separation, revealing two groups with significant differences in intestinal parameters (villus height-to-crypt depth ratio, intestinal absorptive area, duodenal maltase activity, butyric, propionic and total volatile fatty acid concentrations) and performance outcomes (body weight at 15 days and average daily gain). Supervised models were subsequently applied as interpretative tools to assess variable relevance, with Random Forest achieving high internal consistency. SHAP analysis indicated that intestinal morphology, enzymatic activity, and microbial metabolites (particularly total volatile fatty acids, propionate, and butyrate) were most strongly associated with cluster classification. These findings highlight coordinated patterns between intestinal function and growth during the early post-weaning period and suggest that such biomarkers may represent potential targets to be explored in future nutritional strategies. Overall, this study demonstrates the potential of integrating unsupervised explainable machine learning approaches into animal science research for exploratory analysis and hypothesis generation.