The stability paradox: Why high prediction accuracy does not guarantee reliable feature importance in psychiatric research.

Assessing marine ecosystem risks through unsupervised methods

Novel unsupervised Bayesian method for Near Real-Time forest loss detection using Sentinel-1 SAR time series: Assessment over sampled deforestation events in Amazonia and the Cerrado

Subsurface defect detection of concrete-filled steel tubular (CFST) structure based on a two-stage unsupervised learning method

Unsupervised fault diagnosis method for rolling bearings based on federated universal domain adaptation

Hybrid unsupervised methods and inject-multiply morphological features for mapping wildfire burned areas with multi-spectral satellite data

Background Type 2 diabetes mellitus (T2DM) is a major global public health issue, with a particularly high prevalence in China, especially among older men. Obesity, dietary habits, and metabolic risk factors are key contributors to the development of T2DM. However, research on the relationship between dietary patterns, obesity, and T2DM in elderly Chinese men remains limited. Objective: This study aims to examine the links between obesity, dietary habits, blood pressure, and the risk of developing T2DM in elderly Chinese men. We utilize unsupervised machine learning methods along with SHAP-based model interpretation to identify significant lifestyle and metabolic factors associated with T2DM risk. Methods A cross-sectional study was conducted with 982 participants aged 60 years and older from community health centers in Heze City, China. Unsupervised machine learning methods (UMAP) were used to identify dietary patterns, and supervised machine learning with SHAP was applied to evaluate the importance of obesity, dietary patterns, and lifestyle factors on T2DM risk. Logistic regression analyses were performed to investigate the associations between obesity, dietary habits, blood pressure, and T2DM risk. Sensitivity analyses were performed to verify the robustness of the findings. Results Four distinct dietary patterns were identified: “high-fiber nutrient-dense,” “staple–protein,” “seafood-eggs,” and “sugary and processed foods.” The prevalence of newly diagnosed T2DM in males was 48.37%. Obesity was inversely associated with T2DM risk across all models (odds ratios: 0.272–0.278, all P < 0.05). Compared with the high-fiber nutrient-dense pattern, adherence to the staple–protein, seafood–eggs, and sugary and processed foods patterns was significantly associated with increased obesity and T2DM risk (all P < 0.01). Shapley Additive Explanations (SHAP) analysis highlighted dietary behaviors, total energy intake, and physical activity as major contributors to T2DM prediction. Sensitivity analyses confirmed the robustness of these associations, independent of total caloric intake and BMI. Conclusion In this population of elderly Chinese males, unhealthy dietary patterns are positively associated with obesity and T2DM risk, whereas obesity itself showed an inverse relationship with T2DM. These findings underscore the importance of promoting nutrient-dense diets and targeted lifestyle interventions to reduce T2DM risk in this population.

In intelligent sports education, current action quality assessment (AQA) methods face significant limitations: regression-based methods are heavily dependent on high-quality annotated data, while unsupervised methods lack sufficient accuracy and degrade performance when handling long-duration sequences. To address these challenges, this paper introduces a novel indirect scoring method integrating action anomaly detection with a Quick Action Quality Assessment (QAQA) algorithm. In this method, the proposed anomaly detection module dynamically adjusts action quality scores by identifying and analyzing acceleration outliers between frames, effectively improving the robustness and accuracy of sports AQA. Moreover, the QAQA algorithm utilizes a multi-resolution approach, including coarsening, projection, and refinement, to significantly reduce computational complexity to O(n), alleviating the computational burden typically associated with long sequence analyses. Experimental results demonstrate that our method outperforms traditional methods in execution efficiency and scoring accuracy. The proposed system improves algorithmic performance and effectively contributes to intelligent sports training and education.

With the increasing globalization of supply chains, ensuring food safety has become more complex, necessitating advanced approaches for risk assessment. This study aims to review the transformative role of machine learning (ML) and deep learning (DL) in enabling intelligent food safety management by efficiently analyzing high-quality and nonlinear data. We systematically summarize recent advances in the application of ML and DL, focusing on key areas such as biotoxin detection, heavy metal contamination, analysis of pesticide and veterinary drug residues, and microbial risk prediction. While traditional algorithms including support vector machines and random forests demonstrate strong performance in classification and risk evaluation, unsupervised methods such as K-means and hierarchical cluster analysis facilitate pattern recognition in unlabeled datasets. Furthermore, novel DL architectures, such as convolutional neural networks, recurrent neural networks, and transformers, enable automated feature extraction and multimodal data integration, substantially improving detection accuracy and efficiency. In conclusion, we recommend future work to emphasize model interpretability, multi-modal data fusion, and integration into HACCP systems, thereby supporting intelligent, interpretable, and real-time food safety management.

Machine learning models trained on computed tomography (CT) images are highly sensitive to variations in imaging acquisition parameters. Even subtle inconsistencies, often unnoticeable to human radiologists, can significantly degrade model accuracy. In clinical practice, datasets frequently exhibit heterogeneity due to variations in imaging protocols and scanner characteristics, which makes associated metadata a valuable but often underutilized resource for identifying sources of bias. To address this, we propose a novel unsupervised method that systematically identifies confounding and potentially confounding factors embedded in metadata. The key strengths of our method include automated detection of influential metadata attributes, minimal reliance on manual input, and the capability to proactively flag variables that could induce model drift post-deployment. Empirical evaluation in two distinct CT datasets demonstrates that controlling for factors identified by our method drastically improves model performance, increasing classification accuracy by 5 to 15% compared to datasets where these factors remain uncontrolled. These comparative results underscore the potential of our approach to substantially improve the robustness, consistency, and clinical applicability of radiomic machine learning models.

Abstract Microtextural and chemical data from minerals in igneous rocks are critical for unpacking and developing understanding of processes in magmatic systems. Recent advancements leveraging unsupervised machine learning methods offer novel approaches for phase classification without requiring prior knowledge of phases or chemistry. The present work expands on these methods through a probabilistic framework to enable a high‐throughput approach, which can result in significant improvements in segmentation of noisy EDS data collected during SEM mapping of rock specimens. We demonstrate that linear matrix decomposition methods, such as principal component analysis (PCA), can be applied to examine the chemical zonation of mineral phases without prior knowledge of their chemistry. To calibrate phase chemistry, we introduce a Bayesian Markov Chain Monte Carlo (MCMC) approach that scales high spatial resolution energy dispersive spectroscopy (EDS) data to high precision and accuracy electron probe microanalysis (EPMA) profiles. We use this technique to derive high quality and reproducible chemical data sets across large spatial fields. The proposed method couples textural and chemical observations that allow petrologists to better interpret magmatic systems and understand crustal processes.

Abstract Conventional link prediction methods mainly aim to estimate pairwise relationships between nodes in graph structures, typically addressing single-type interactions. However, real-world complex systems often exhibit high-order group relationships that extend beyond binary interactions. For instance, a research paper is often co-authored by multiple researchers. To address the loss of high-order structural information in traditional graph models for representing multivariate interactions, we propose HP2PH, a novel hyperlink prediction method based on 2-head preferential hypergraph weighted random walk with restart. Firstly, considering the uncertainty of hyperedge cardinalities in non-uniform hypergraphs, we introduce a preferential hypergraph weighted random walk with restart strategy, called P-HWRWR. This strategy fully exploits the high-order topological properties of hypergraphs, and jointly optimizes the random walking sub-paths from node to hyperedge and from hyperedge to node by assigning weights to both the hyperedges and nodes encountered by the random walker. Subsequently, an unsupervised hyperlink prediction method based on the two-head preferential hypergraph weighted random walk with restart is proposed. This approach searches for potential member nodes within new hyperedges from different directions, ensuring efficient predicting multivariate interactions with relatively low time complexity. Finally, through extensive experimental verification and analysis on 10 non-uniform hypergraph datasets and 5 uniform hypergraph datasets, it is demonstrated that HP2PH achieves improvements of 0.8% to 137.2% in the AFS metric and 0.8% to 412.1% in the HPA metric on non-uniform datasets and achieves improvement of 94.4% to 202.5% in the AFS metric on uniform hypergraph datasets compared baselines. These experiments substantiate the superiority and operational viability of the developed method when predicting hyperlinks.

Prostate cancer is a prevalent malignant tumor in men, and accurate diagnosis and personalized treatment rely on multimodal imaging, such as MRI and TRUS. However, differences in imaging mechanisms and prostate deformation due to ultrasound probe compression pose significant challenges for high-quality registration between the two modalities. In this study, we propose a label-aware weakly supervised diffusion model for MRI-TRUS multimodal image registration. First, we align label centroid positions by maximizing the Dice coefficient to correct initial biases. Second, we combine label supervision with a diffusion model to generate high-quality deformation fields. Finally, we incorporate a feature-guided module to better preserve edge structures and improve registration smoothness. Experiments conducted on the µ-RegPro dataset demonstrate that our method outperforms current state-of-the-art (SOTA) approaches across multiple evaluation metrics. Specifically, it achieves a Dice coefficient of 0.880 and reduces the target registration error (TRE) to 0.940, significantly surpassing unsupervised methods such as VoxelMorph, FSDiffReg, and supervised methods like LocalNet and AutoFuse. The results show that preliminary label centroid alignment effectively enhances the performance of the diffusion-based deformation registration model, reducing the TRE from 3.084 to 0.940. The ablation study demonstrates that the feature-guided diffusion module effectively suppresses deformation field folding, while the label-aware module enhances label alignment. When combined, the proposed framework achieves a favorable balance, substantially improving registration accuracy (Dice = 0.880, TRE = 0.940) with reduced folding (|J|≤0 = 0.134). This method exhibits strong robustness and generalizability in handling large deformations in target regions while preserving details in nontarget regions. The proposed label-aware weakly supervised diffusion model enables accurate and efficient MRI-TRUS multimodal image registration, offering strong potential for clinical applications such as prostate cancer diagnosis, targeted biopsy, and image-guided navigation.

Purpose This study aims to investigate the key drivers of bank stability in Vietnam's emerging economy, offering a robust, data-driven framework that integrates advanced machine learning (ML), regularization techniques and explainable artificial intelligence (XAI) to address challenges in financial risk modeling and regulatory transparency. Design/methodology/approach Using bank-level data from 2010 to 2023, this study employs Ridge, Lasso and Elastic Net regression to manage multicollinearity and identify relevant predictors. Gradient boosting with ridge regularization, optimized via particle swarm optimization, achieves superior predictive accuracy (R2 = 96.03%). SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) are applied to interpret the model's outputs, revealing both global and local effects of explanatory variables. Findings The model attains a strong explanatory power (R2 = 96.03%), affirming the validity of the hybrid ML-XAI approach. Core drivers of banking stability – lag_ZSCORE, foreign bank presence, return on equity, equity ratio and macroeconomic factors – are consistently identified by SHAP and LIME. The integration of interpretable artificial intelligence (AI) not only enhances predictive accuracy but also delivers actionable insights, offering meaningful guidance for financial stability in emerging markets. Research limitations/implications While the model demonstrates strong performance within Vietnam's banking sector, its applicability may vary in different regulatory or macroeconomic environments. The study focuses on supervised learning and structured data; future research could explore unsupervised methods or unstructured sources like textual financial disclosures. Additionally, while SHAP and LIME provide interpretability, they do not guarantee causal inference. Nevertheless, the research lays a solid methodological foundation for future cross-country comparisons, particularly in the Association of Southeast Asian Nations, and encourages the integration of XAI into financial stability frameworks globally. Practical implications This study equips financial regulators, central banks and policymakers with interpretable and high-performing tools to monitor and enhance banking stability. By clearly identifying and quantifying key stability drivers, it enables targeted, data-informed interventions and policy adjustments. Banks can also utilize these insights to optimize risk management, capital allocation and strategic planning. The transparent AI methods ensure trust in the decision-support system, promoting broader adoption in regulatory environments that demand both accuracy and explainability. Originality/value This is the first study in Vietnam to fuse ML, regularization and XAI for bank stability analysis. It not only enhances predictive power but also ensures model transparency, crucial for policymaking and Basel III compliance. The methodological innovation offers a replicable blueprint for financial risk assessment in emerging markets.

The symmetry between different representation spaces plays a crucial role in effectively modeling complex multimodal data. To address the challenge of equipment knowledge graphs containing hierarchical relationships that cannot be fully represented in a single space, this study proposes UMEAD, an unsupervised multimodal entity alignment method based on dual-space embeddings. The method simultaneously learns graph embeddings in both Euclidean and hyperbolic spaces, forming a structural symmetry where the Euclidean space captures local regularities and the hyperbolic space models global hierarchies. Their complementarity achieves a balanced and symmetric representation of multimodal knowledge. An adaptive feature fusion strategy is further employed to dynamically weight semantic and visual modalities, enhancing the symmetry and complementarity between different modalities. To reduce reliance on scarce pre-aligned data, pseudo seed instances are generated from multimodal features, and an iterative constraint mechanism progressively enlarges the training set, enabling unsupervised alignment. Experiments on public datasets, including EMMEAD, FB15K-DB15K, and FB15K-YAGO15K, demonstrate that the combination of dual-space embeddings, adaptive fusion, and iterative constraints significantly improves alignment accuracy. In summary, the proposed method reduces dependence on pre-aligned data, strengthens multimodal and structural alignment, and its symmetric embedding and fusion design offers a promising approach for the construction and application of multimodal knowledge graphs in the equipment domain.

Background In any technical field, Non-Technical Skills (NTS) complement the work activity. Communication is one of these essential skills, required in any activity involving social and professional interaction. These communication skills are crucial especially in high-risk industrial environments. This collective and social NTS can be a contributing factor in many workplace accidents. Nowadays, it is becoming increasingly crucial to be able to act on this skill, starting with its diagnosis and objective assessment. Aim This study presents an assessment of this specific NTS through a Virtual Reality (VR) simulation. The VR scenario used involve a team solving of a collective task under time pressure. This virtual environment has been designed to match the anxiety-inducing, high-risk industrial context. Method N = 59 participants were included in the study and divided into 12 groups. Quantitative and qualitative data were collected to assess the overall groups’ performance. This mixed data will be analyzed and presented to compare team communication when performing a time-pressured collaborative task in an immersive environment. First, we will analyze the quantitative performance data using the k-means unsupervised clustering method guided by principal component analysis (PCA). Verbal data were then studied to account for differentiations in the communicative skills mobilized and which will characterize each of the clusters. Results Three distinct clusters emerged, each representing different performance patterns. Analysis of the verbal data revealed that these clusters correspond to varying levels of communication skills. Conclusion Virtual reality simulation is an effective tool for assessing group communication, and can be an effective training tool combined with structured debriefing.

Abstract Chaotic time series pose significant challenges for anomaly detection tasks due to their highly nonlinear characteristics. Traditional methods struggle to accurately simulate their dynamic evolution, effectively handle chaotic noise, and meet the demands of rapid inference, unlabeled datasets, and long-term data persistence, all of which hinder the development of efficient and accurate anomaly detection models. To address these issues, this study proposes an unsupervised anomaly detection method based on Empirical Mode Decomposition (EMD) and Recurrent Convolutional Encoding Attention mechanism (RC-Attention). This method first reconstructs the original sequence from high to low frequencies using an improved EMD to mitigate high-frequency noise interference. Then, it reconstructs the phase space of the decomposed sequence and inputs the multi-dimensional array into the network. To fit the time-dependent relationship, a recurrent convolutional encoding attention mechanism is introduced to mine chaotic sequence features: recurrent convolutional encoding converts long sequences into short sequences with a tower structure and upsampling encoding, retaining key information; the central stationary attention mechanism learns temporal relationships and reduces the tendency to learn abnormal states; and stacked large-kernel convolution learns the correlation between embedding dimensions. Compared to baseline methods, RC-Attention effectively avoids overfitting on abnormal states and significantly improves generalization ability and time-dependent relationship extraction ability. Experiments on Lorenz, Rossler, and energy consumption datasets show that this model can efficiently recover and reconstruct chaotic sequence features, with an average F1 score improvement of 14.9%, validating its effectiveness and superiority.

Abstract. Accurate registration between RGB-D images and point clouds is a critical task for various indoor applications. Estimating the relative pose by aligning the sensor frame with indoor 3D point clouds significantly enhances environmental perception and scene understanding. Existing research primarily focuses on cross-modal feature association through traditional unsupervised methods or supervised learning-based approaches. However, these methods often rely on strong assumptions, such as the availability of an initial pose or substantial overlap between the RGB-D images and the target point clouds. Moreover, the quality of registration is highly sensitive to the density and completeness of the point clouds. To address these limitations, this paper presents a novel coarse-to-fine registration framework with the aid of CAD models. First, a data enhancement process is introduced using the Scan2CAD method to replace functional objects (e.g., chairs and tables) with CAD models, improving semantic and quality consistency. Second, a geometry-aware graph matching is computed to identify regions of interest (ROI) within the point cloud map and estimate the initial pose of the RGBD sensor. Finally, an iterative fine matching using cross-modal is introduced to refine the initial estimated pose. Experimental validation on the ScanNet dataset demonstrates that the proposed framework achieves robust and accurate registration between RGB-D images and 3D point clouds.

Pancreatic ductal adenocarcinoma (PDAC) is a rapidly progressing cancer that responds poorly to immunotherapies. Intratumoral tertiary lymphoid structures (TLS) have been associated with rare long-term PDAC survivors, but the role of TLS in PDAC and their spatial relationships within the context of the broader tumor microenvironment remain unknown. In this study, we report the generation of a spatial multiomic atlas of PDAC tumors and tumor-adjacent lymph nodes from patients treated with combination neoadjuvant immunotherapies. Using machine learning-enabled hematoxylin and eosin image classification models, imaging mass cytometry, and unsupervised gene expression matrix factorization methods for spatial transcriptomics, we characterized cellular states within and adjacent to TLS spanning distinct spatial niches and pathologic responses. Unsupervised learning identified TLS-specific spatial gene expression signatures that are significantly associated with improved survival in patients with PDAC. We identified spatial features of pathologic immune responses, including intratumoral TLS-associated B-cell maturation colocalizing with IgG dissemination and extracellular matrix remodeling. Our findings offer insights into the cellular and molecular landscape of TLS in PDACs during immunotherapy treatment.

A Level I trauma center used machine learning algorithms to identify risk factors and patterns in falls among older adults, which constitute our greatest burden of traumatic admissions. A retrospective analysis was conducted on 2,391 ground-level fall trauma admissions from 2017-2022 including variables related to demographics, and weather conditions at admission. Supervised learning models were developed to predict older adult vs younger counterpart falls. In this machine learning modality, we generated a Decision Tree, a Support Vector Machine Classifier Algorithm, and a Logistic Regression Model. Unsupervised learning methods uncover patterns or groupings in the dataset of older adult ground-level falls, which consists of 1,742 records from 2017-2022 trauma admissions including comorbidity variables. Unsupervised learning methods of Principal Components Analysis, Hierarchical Clustering, and Market Basket Analysis were employed. All three supervised models found the female sex as an important variable in predicting older adult falls. Unsupervised learning identified discernible patterns and groupings, revealing that certain weather variables are associated with falls. These machine learning modalities can shed light on what may be important risk factors for older adult falls and can help to target awareness and outreach.