Search Results (3,734)

This paper addresses the problem of nonparametric regression for functional time series, a setting complicated by the infinite-dimensional nature of the covariates, temporal dependence, and potential for outliers. We propose a new robust estimator that combines three powerful ideas: (i) k-nearest neighbors (kNN) for adaptive localization in the functional space; (ii) local linear smoothing to reduce bias; and (iii) $\mathcal{M}$-estimation to ensure resilience against atypical observations. The key theoretical contribution establishes the almost-complete convergence of the proposed estimator under mild conditions that account for the functional geometry, weak dependence (via quasi-association), and robustness constraints. The obtained rate of convergence explicitly reveals the interplay between the functional concentration, dependence strength, and local smoothness of the model. A simulation study demonstrates that this method offers superior stability and predictive accuracy compared to classical alternatives, particularly under heavy-tailed errors and data contamination. The practical relevance of the approach is further illustrated through a one-step-ahead prediction application to a real-world environmental dataset of hourly NOx measurements. Full article

Light Detection and Ranging (LiDAR) three-dimensional (3D) object detection degrades under point sparsity, outliers, coordinate noise, and calibration drift, yet detector evaluation remains largely limited to clean benchmarks. This study focuses on sensing robustness rather than detector redesign. We introduce Bounded Graph Conditioning (BGC)—a deterministic pre-voxelization front-end that applies k-nearest-neighbor (kNN) neighborhood averaging with bounded residual correction upstream of an unchanged detector backbone. BGC is evaluated together with a reproducible sensor-degradation stress protocol and a risk-constrained operating-boundary analysis. Experiments on KITTI with PointPillars, SECOND, and Voxel R-CNN show that BGC most clearly improves retained detection quality and feasible operating coverage under strong noise and strong outlier stress; gains under other degradation types are smaller and backbone-dependent. In the primary score-level box-disjoint calibration/test evaluation on SECOND, maximum feasible coverage at a target risk bound of $0.2$ improves from $0.0754$ to $0.1374$ under strong noise ($\sigma =0.10$ m) and from $0.1323$ to $0.1591$ under strong outliers ($p=0.10$); a cross-backbone check on Voxel R-CNN confirms the same direction ($0.1860\to 0.2864$). Comparison with traditional filtering (SOR and ROR) reveals complementary strengths across fault types. A range-adaptive BGC variant that adjusts parameters per distance bin further improves performance under mixed unknown faults, spherical-coordinate noise, and on a dataset-matched nuScenes validation (adaptive BGC mAP/NDS: $0.2687/0.4493$ vs. baseline $0.2471/0.3846$ under strong noise). Severe translation drift collapses all configurations to full rejection, exposing an explicit sensing boundary beyond the reach of local conditioning. These results support BGC as a practical sensor-side robustness enhancement under the studied degradation protocol, with conditional rather than universal applicability across backbones and fault types. Full article

Frequency response analysis (FRA) is recognized as an effective method in power transformer winding fault diagnosis. However, the traditional numerical index methods focus on the overall features of FRA curves, making it difficult to capture subtle deformations in transformer windings. Similarly, existing digital image processing methods rely on a single feature or a simple feature combination without adaptive fusion. These methods ignore differences in the data distributions of features, leading to feature mismatch, the loss of sensitive fault information, and lower diagnostic accuracy. To solve this problem, a novel adaptive multiple-image-feature fusion method for transformer winding fault diagnosis is proposed. First, a multi-dimensional feature space combining image pixel matrix similarity, morphological features, and image texture features is built to decode the difference in fault of FRA images. Second, the multiple kernel learning (MKL) framework is used to dynamically adjust the fusion weights of different kernels to make features compatible and remove redundant information. Finally, comparative and ablation experiments show that the proposed method outperforms the traditional methods in identifying different types and levels of faults. The method achieves over 99% accuracy in fault type identification across SVM, KNN, and RF classifiers. For radial deformation (RD) severity prediction, the accuracy of the proposed model is 93.37% with SVM and 94.85% with KNN, outperforming the full-feature concatenation method. These results confirm the method’s robustness and diagnostic precision. Full article

Clinical prediction from electronic health records (EHRs) is complicated by pervasive missingness and label scarcity, which make performance sensitive to the match between data conditions and pipeline choice. Choosing the best pipeline for a new incomplete dataset still requires costly trial-and-error. We cast this as an algorithm selection problem and address two bottlenecks—instance scarcity and distance quality—that have so far prevented meta-learning from reaching clinical settings. Graph neural networks offer diverse strategies (patient similarity networks, bipartite imputation graphs, attention-driven feature interaction), yet no single architecture dominates across missingness patterns, and selecting the best pipeline for a new dataset remains a trial-and-error approach. Formal algorithm selection could automate this choice but requires many characterized meta-instances—more than clinical settings typically provide. We propose two solutions: (1) constructive instance augmentation, applying controlled quality perturbations (MCAR and MNAR missingness injection, label trimming) to 20 base EHR datasets to expand the meta-knowledge base to 83 characterized meta-instances, each described by a 10-dimensional missingness fingerprint, without additional model training; and (2) dynamic-supervised metric learning, using differential evolution to optimize fingerprint feature weights so that static distances preserve method-performance similarity captured by dynamic fingerprints, which require model sweeps and are unavailable at deployment. Under base-dataset-level leave-one-dataset-out cross-validation over 21 pipelines, the resulting metric-learned kNN recommender attains the highest win rate (20.5%) among non-oracle strategies on the augmented store, selecting the correct pipeline more often than any fixed default. At deployment, the recommender needs only the 10-dimensional static fingerprint with pre-learned weights; no sweep data is required for new datasets. Cross-domain evaluation on 25 external subsets (colorectal cancer, kidney disease, MIMIC-IV) demonstrates framework modularity: when the fingerprint module is adapted (standard meta-features in place of the missingness-specific set), the recommender achieves regret of 0.025 (55% below random selection). Full article

Unlike reactive systems, mechanical ventilation controlled by CO₂ concentration operates at a target efficiency that dynamically increases whenever the target CO₂ level is exceeded. This approach eliminates the typical ‘dead-time’ and prevents air quality degradation by ensuring the system adjusts its performance immediately in response to concentration changes. In this work, the study focuses on the development and evaluation of data-driven predictive models for near-term indoor CO₂ forecasting that can be integrated into pre-occupancy ventilation strategies, rather than designing a complete control scheme. Experimental data were collected over four months in a 48 m² smart laboratory configured as an open-plan office, where a heterogeneous IoT sensing architecture logged synchronized time-series measurements of CO₂ and microclimate variables (temperature, relative humidity, PM_2.5, TVOCs), together with acoustic noise levels and appliance-level energy consumption used as indirect occupancy-related signals. Raw telemetry was transformed into a 22-feature state vector using a structured feature engineering method incorporating z-score standardization, cyclic time encodings, multi-horizon CO₂ lags, rolling statistics, momentum features, and non-linear interactions to represent temporal autocorrelation and daily periodicity. The study benchmarks multiple regression paradigms, including simple baselines and ensemble methods, and found that an automated multi-level stacked ensemble achieved the highest predictive fidelity for short-term forecasting, with an Mean Absolute Error (MAE) of 32.97 ppm across an observed CO₂ range of 403–2305 ppm, representing improvements of approximately 24% and 43% over Linear Regression and K-Nearest Neighbors (KNN), respectively. Temporal diagnostics showed strong phase alignment with observed CO₂ rises during occupancy transitions and statistically reliable prediction intervals. Five-fold walk-forward cross-validation confirmed the temporal stability of these results, with top models achieving consistent R² values of 0.93–0.95 across Folds 2–5. These results demonstrate that, within a single-room university laboratory setting, historical sensor data from low-cost IoT devices can support accurate short-term CO₂ forecasting, providing a predictive layer that could support future proactive ventilation scheduling aimed at reducing CO₂ lag at the start of occupancy while avoiding unnecessary ventilation runtime. Generalization to other building types and occupancy profiles requires further validation. Full article

Impulsive alpha (α)-stable noise, characterized by heavy tails and intense outliers, is a key ingredient in simulating financial, medical, seismic, and digital communication technologies. It poses versatile challenges to conventional machine learning (ML) algorithms in predicting noise parameters for multidisciplinary artificial intelligence (AI)-embedded devices. In this study, we adopted a two-phase methodology to investigate the complexity and performance of supervised ML algorithms while classifying impulsive noise parameters. We generated synthetic datasets of α-stable noise distributions for experimentation in a controlled environment. It was followed by experimental evaluation to derive the complexity and performance of ML classifiers—k-nearest neighbors (KNN), Support Vector Machine (SVM), Naïve Bayes (NB), Decision Tree (DT), and Random Forest (RF). Moreover, we employed a very high channel noise level of −15 dB in the test datasets to ensure that the derived analysis applies to real-world devices. The results demonstrate the high performance of DT and RF in structured binary classification of the α regime and the sign of skewness, while incurring satisfactory computational costs. However, SVM and kNN are comparatively more robust for multi-class classification, albeit with higher memory and training costs. On the contrary, NB fails to address the skewed and impulsive behavior of α-stable noise. We observed that even the most effective classifiers struggle to achieve perfect accuracy in multi-class classification. Overall, the experimental results reveal significant trade-off relationships between the complexity and performance of ML classifiers. Conclusively, simple models are well-suited for coarse-grained tasks, such as α-approximation and sign-of-skewness classification. In contrast, sophisticated models can be deployed to predict noise parameters to some extent. Our study provides a clear set of trade-offs for future applied AI devices that address adversarial and impulsive noise. Full article

Background: As muscles atrophy, myocytes are replaced by fat and muscle strength diminishes, increasing thoracic supine kyphosis. Here, we investigate the relationship between muscle fat percentage (muscle–fat%) and thoracic supine kyphosis on CT. Methods: Thoracic Cobb angle was measured on supine CT scans from the AtlasDataset by four observers ($n=533$). Nine muscles were manually labeled on 100 scans (manual cohort). An nnU-Net model was trained on 80 cases with internal validation on 20 cases, then applied to segment the remaining 433 AtlasDataset scans (automated cohort). External segmentation benchmarking was performed on 30 cases from a separate open-source dataset. Associations between supine thoracic curvature and muscle–fat% were evaluated only in AtlasDataset. Results: Manual supine thoracic Cobb angle measurements demonstrated good inter-observer reproducibility (ICC(2,k) = 0.98) with a mean across-rater per-case SD of 3.4${}^{\circ }$. The nnU-Net achieved Dice scores >0.93 across all nine muscle groups on internal and external segmentation benchmarking. For both manual and automated cohorts, thoracic supine kyphosis correlated with muscle–fat% in the paraspinal (r = 0.35, 0.42), quadratus lumborum (r = 0.29, 0.33), vastus (r = 0.38, 0.32), psoas (r = 0.21, 0.23) and latissimus dorsi (r = 0.21, 0.17) muscles. Conclusions: Automated measurement of trunk muscle–fat% provides a reproducible imaging biomarker correlated with thoracic supine kyphosis on CT. Identifying fatty atrophy of core muscles may help identify potential targets for interventions in hyperkyphotic patients. Full article

Point cloud segmentation is a core technology in remote sensing, enabling the extraction of rich semantic information from complex scenes. Existing methods struggle with semantic inconsistency across multiple heterogeneous datasets in complex urban environments. To address semantic inconsistencies, we propose SemAlign3D, a novel multimodal framework for point cloud segmentation that combines learnable class prompts with a multi-scale feature attention module. We integrate five large-scale datasets (SensatUrban, STPLS3D, WHU3D, SemanticKITTI, Semantic3D) to construct a unified training framework, ensuring label consistency by recalibrating semantic labels. The learnable class prompt mechanism dynamically adapts to dataset-specific semantics, enhancing the semantic consistency across multiple datasets of point cloud segmentation. Additionally, the Multi-scale K-Nearest Neighbor Feature Attention Enhancement module integrates local and global features, improving semantic discriminability in complex scenes. Within a single unified training framework, our method effectively aligns semantic labels from multiple heterogeneous datasets, achieving gains of +1.61% mIoU on WHU3D and +0.98% mIoU on SemanticKITTI. These results demonstrate the effectiveness of our framework in improving semantic consistency and robustness across heterogeneous point cloud datasets. Full article

Corporate risk prediction is a central problem in financial analysis and corporate risk management. This study proposes a functional approach in which firms are represented through multivariate financial trajectories constructed from retrospective windows of accounting indicators, over which a similarity measure is defined and incorporated into a k-nearest neighbors classifier. The target variable is derived from administrative records, combining reporting discontinuity and firm administrative status as a proxy for financial distress. The empirical application is conducted using data from firms in the tourism sector in Colombia and is evaluated through stratified cross-validation. The results show that the trajectory-based representation captures gradual patterns of financial deterioration and improves the performance of k-NN relative to its static variable counterpart. In addition, the approach enhances interpretability by enabling the identification of historically comparable firms and the analysis of the financial dimensions that explain their similarity. Overall, the model provides a complementary perspective for corporate risk analysis based on the comparison of financial trajectories. Full article

Reliable short-term prediction of bus travel time on signalized urban arterials is essential for improving service reliability and may provide a useful forecasting basis for prediction-informed transit signal priority (TSP) and arterial coordination applications. However, bus operations on urban arterials are highly variable due to stop dwell times, signal delays, and interactions with mixed traffic, leading to nonlinear and nonstationary travel time patterns with strong spatiotemporal dependence. This study proposes a hybrid KNN–ConvLSTM framework for short-term arterial bus travel time prediction using real-world field data. A K-nearest neighbors (KNNs) module is first employed to retrieve historical operation sequences that are most similar to the current corridor state, thereby reducing interference from mismatched traffic regimes and improving robustness. Smart-card (IC card) transaction data are incorporated as demand-related features to represent passenger activity and its impact on dwell time and travel time variability. The selected sequences are then organized into a corridor-ordered spatiotemporal representation and further refined by lightweight temporal enhancement operations, including relevance gating, multi-scale aggregation, adaptive feature fusion, and residual enhancement, before being fed into the convolutional long short-term memory (ConvLSTM) predictor. The proposed approach is evaluated using weekday service-hour data extracted from 30 days of real-world bus operation records collected from a typical urban arterial corridor in Changchun, China, and is compared with several benchmark models, including ARIMA, KNN, LSTM, CNN, ConvLSTM, Transformer, and DCRNN. The results indicate that the proposed KNN–ConvLSTM framework achieves an MAE of 40.1 s, an RMSE of 55.8 s, a SMAPE of 10.7%, and an R² of 0.878, outperforming all benchmark models. Specifically, compared with the Transformer baseline, the proposed framework reduces MAE by 1.5%, RMSE by 5.1%, and SMAPE by 7.0%, while increasing R² by 0.014. Compared with the DCRNN baseline, it reduces MAE by 10.7%, RMSE by 1.9%, and SMAPE by 2.7%, while increasing R² by 0.008. These findings demonstrate that similarity-aware retrieval combined with spatiotemporal deep learning can substantially enhance short-term bus travel time prediction on signalized urban arterials. More accurate short-term forecasts may support prediction-informed transit signal priority and arterial coordination by providing more reliable downstream arrival-time estimates. However, the generalizability of the reported results is still constrained by the relatively short 30-day observation period and the single-corridor case setting, and the operational and environmental effects of downstream applications remain to be validated through dedicated closed-loop control evaluation in future work. Full article

Large-scale chest X-ray (CXR) datasets often rely on report-derived or weak labels, introducing missing and incorrect annotations that can degrade downstream models and limit trust. We study training-free mislabel detection in multi-label CXRs by scoring neighborhood label consistency in a fixed embedding space. Using the NIH Chest X-ray Kaggle sample (5606 CXRs), we extract intermediate CoAtNet features and obtain 64-dimensional embeddings with a frozen CoAtNet backbone and a lightweight refinement head. On top of these embeddings, we compare kNN consistency baselines with distance weighting and label-set similarity against LPV-DW-CS, clustered prototype voting weighted by distance and cluster support. We evaluate three synthetic label-noise regimes with review budgets matched to the corruption rate: random single-label (5% and 20%), boundary-noise (20% corruption within the lowest-density 20% subset), and disjoint-label replacement (20% within that subset). LPV-DW-CS yields the highest downstream macro-AUROC after filtering top-ranked samples (up to 0.8860), while kNN variants achieve higher Recall@budget at the same review rates (up to 99.44%). An image-only expert Likert review of top-ranked real samples finds substantial label-set inconsistencies (54.1% for LPV-DW-CS-280-A; 60.5% for KNN-DW-LSS), supporting neighborhood-consistency ranking as a practical, training-free tool for targeted dataset auditing. Full article

The combination of Fourier-transform infrared (FTIR) spectroscopy and machine learning gives a promising result in wheat variety classification. This study aimed to evaluate the contributions of distinct spectral regions and their combinations to classification performance. Out of the full raw spectra of four bread wheat varieties, namely Altindane, Cavus, Flamura-85, and Nevzatbey, 15 spectral datasets were prepared. Artificial Neural Networks (ANN), Support Vector Machines (SVM), Random Forest (RF), and K-Nearest Neighbor (KNN) models were trained and analyzed. The highest classification performance was obtained using spectral regions associated with protein and lipid bands. The highest average accuracy of 0.9895 was shown by the SVM model, while the ANN produced comparable results with lower variability. Additionally, Variable Importance in Projection (VIP) analysis identified the most influential spectral bands in the protein (Amide II, ~1542 cm⁻¹) and carbonyl (1744–1715 cm⁻¹) regions. These findings indicate that classification is driven by chemically meaningful features rather than purely statistical patterns. The approach followed in this study provides an insight that, in FTIR-based classification, when rigorously evaluated using nested cross-validation, spectral region selection can outweigh model complexity. This approach demonstrates strong potential for rapid and non-destructive assessment, especially for real-time applications in grain processing and automated sorting systems. Full article

Blade impact faults degrade power generation quality, if not detected in time, may lead to turbine malfunction or even complete failure. Moreover, the accuracy of blade impact fault detection in tidal current turbine (TCT) is significantly affected by variations in flow velocity and tidal flow period. To solve this problem, a self-adaptive detection method based on stator current signals and k-nearest neighbor-multiplicative score (KNN-MS) is proposed. The method first employs the KNN algorithm to characterize local feature distributions. Then, robustness under unstable flow conditions is improved through variance-based weighting. Finally, a cumulative multiplicative scoring mechanism is proposed to amplify and quantify fault-related anomaly indicators. The experimental results show that the proposed method achieves high diagnostic accuracy and stability across steady, periodic, and variable-period flow scenarios. Full article

Background/Objectives: Cardiovascular diseases are the leading cause of death worldwide. Therefore, early diagnosis and treatment of these diseases are of critical importance. Stethoscopes are the easiest and fastest medical devices for the initial diagnosis of cardiovascular diseases. However, interpreting heart sounds requires considerable expertise. The use of artificial intelligence in healthcare for decision support has increased and become popular recently. Methods: The popular 2016 PhysioNet/CinC Challenge dataset, consisting of phonocardiogram (PCG) signals, was used to implement the proposed approach. Spectrogram and continuous wavelet transform (CWT) images of the PCG signals were first generated. This increased the distinguishability of the data in terms of both time and frequency components. These two-input images were tested on the developed Dual Time–Frequency Swin Transformer–ConvNeXt Attention Network (DTF-STCANet) model. To further improve classification accuracy, the Weighted KNN algorithm was preferred during the classification phase. Results: With the proposed approach, a 99.29% classification accuracy was achieved. Performance was compared with other state-of-the-art models. Conclusions: The proposed approach, through the integration of PCG signals with artificial intelligence, further strengthens the concept of early diagnosis of heart disease. Full article

The mechanical behavior of aqueduct structures exhibits highly complex characteristics during prestress tensioning, making it difficult for the traditional double-control method to accurately predict and real-time control the key stresses. To improve the construction safety of prestressed tensioning and the prediction accuracy of structural prestress responses, this study develops a rapid structural mechanical property prediction method based on machine learning. Taking prestressed aqueducts as the research object, a system of “finite element simulation—sample generation—machine learning prediction” is established. Firstly, multiple groups of tensioning parameter combinations are designed via Latin hypercube sampling, and the stress responses are obtained through finite element analysis to form a high-quality training sample library. Subsequently, critical structural features are extracted based on mesh reconstruction, and stress prediction models are established using the K-Nearest Neighbors (KNN) and Random Forest algorithms respectively; the prediction performance of both models is compared and validated against finite element simulation results. Furthermore, the prediction outputs of the optimal machine learning model were used to analyze the stress distribution and potential stress concentration issues of the structure during the tensioning process. The comparative analysis results indicate that the Random Forest model performs best in terms of stress prediction accuracy and stability, and its prediction results are highly consistent with those of the finite element method. Compared with traditional finite element condition analysis, the machine learning model can complete multi-condition stress prediction in a shorter time. Leveraging its high-efficiency prediction capability, local high-stress areas of the structure in the tensioning construction scheme can be identified, thereby providing effective optimization schemes to improve the stress distribution. The mechanical response prediction method for the prestress tensioning process of aqueducts, with machine learning as the core, constructed in this paper realizes the rapid and reliable prediction of key stresses throughout the entire prestress tensioning process. This method can be applied to assist in optimizing tensioning construction schemes and construction monitoring, providing a practical technical solution for safety control of aqueduct structures during the prestress construction stage. Full article