Search Results (1,650)

Expectation Maximization (EM) iterations for Gaussian mixture models (GMMs) are highly sensitive to initial parameters, which calls for developing robust initialization methods. For multidimensional GMMs, the problem is even more severe than for univariate GMMs because of the larger number of parameters. For univariate or low-dimensional GMMs, several studies on their initialization have appeared in the literature, whereas research on initializing multivariate, high-dimensional GMMs remains limited. In this study, we compare several initializations for Multivariate Diagonal Gaussian Mixture Models (MDGMMs). In our study, we have included methods already used in the literature for initializing MDGMMs: Hierarchical Clustering, K-means, and random initialization. We have also used a new method, namely, Ensemble Clustering. A review of the existing literature suggests that Ensemble Clustering has not been used previously as an initialization strategy for MDGMM. Several metrics were used to evaluate the clustering quality. Our study demonstrates that Ensemble Clustering, while computationally intensive, is competitive with other methods for initializing MDGMMs. Full article

To meet the requirements of precise temperature regulation in solar greenhouses, traditional machine learning algorithms often suffer from poor adaptability, high energy consumption, and difficulties in integrating agronomic expertise. This study developed an intelligent greenhouse temperature regulation framework based on Model Predictive Control (MPC). The core components of the framework include: (1) an expert-experience-based simulator using a Sparrow Search Algorithm-optimized Random Forest (SSA-RF) model to digitize the temperature management strategies of high-yield farmers into dynamic reference trajectories and (2) a hybrid prediction model (CNN-BiLSTM-Attention) combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise-Permutation Entropy (CEEMDAN-PE) denoising with a Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Attention mechanism to achieve high-precision multi-step temperature forecasting. Validation in a cucumber solar greenhouse demonstrated that the SSA-RF model achieved an R² of 0.976 on the test set, showing a significant improvement over the traditional RF model. Compared to the conventional LSTM model, the hybrid prediction model reduced the RMSE to 0.642 and 0.947 for 15 min and 30 min predictions, respectively, with a maximum R² of 0.994 and excellent generalization capabilities. Finally, these two components were theoretically integrated into an MPC-oriented decision framework. The framework describes how expert reference trajectories, multi-step predictions, actuator constraints, and control increments can be combined in a receding-horizon optimization problem. Since online actuator control data were not available, the MPC module was formulated as a theoretical decision framework rather than a fully validated closed-loop controller. This study provides a modelling basis and technical path for future real-time greenhouse temperature control. Full article

Reliable quantification of compound-drought risk under coupled climate change and land-use dynamics remains a critical challenge. This study employed the Ganjiang River Basin (80,948 km²) as a representative subtropical monsoon catchment, integrating the SWAT hydrological model, CMIP6 multi-model ensembles, the PLUS land-use simulation model, C-vine Copula joint probability analysis, Budyko elasticity attribution, and XGBoost–SHAP decomposition to assess multi-dimensional drought evolution under SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios (2030–2070). The calibrated SWAT model exhibited robust performance (R² = 0.92/0.90; NSE = 0.92/0.89), while the PLUS model (accuracy = 93.6%) projected progressive forest decline (15–19%) with concomitant cropland expansion under escalating emissions. Drought characteristics extracted via run theory from SPEI, SRI, and SSMI revealed a shift toward high-frequency, short-duration events under SSP1-2.6 and pronounced nonlinear amplification under SSP5-8.5. C-vine Copula analysis demonstrated a fundamental transition from Gaussian to Gumbel dependency structures under SSP5-8.5, with three-dimensional AND probability escalating from 0.44 to 0.70 (+59%). Budyko elasticity attribution identified climatic forcing as the dominant driver of runoff variability (94.7–108.9%), while land-use contributions exhibited scenario-dependent sign reversal: forest conservation under SSP1-2.6 suppressed runoff (−8.9%) through enhanced evapotranspiration, whereas forest degradation under SSP5-8.5 amplified runoff (+2.1%) via diminished water retention. XGBoost–SHAP independently corroborated these findings (climate: 95–97%). These results underscore the dominance of climatic forcing in governing drought variability, highlight forest conservation as a cost-effective nature-based mitigation strategy, and emphasize the necessity of multi-index monitoring frameworks for compound-drought risk management. Full article

Machine learning models for medical image analysis are vulnerable to hidden confounders, which can compromise generalization and clinical reliability. Existing detection strategies typically require explicit knowledge or labels of the confounder, which are often unavailable. In this work, we propose an ensemble-based framework to detect potential confounder-driven learning without explicitly defining the confounders, but only which samples might be affected. Our approach leverages the variability of model performance across ensembles to identify signatures of shortcut learning. Shortcut learning occurs when a model uses non-robust features or correlations rather than learning the true underlying task, and it is often observed when confounders are present. We generate controlled dataset variants with increasing confounding levels and analyze distributions of AUC (area under the ROC curve) scores across training, validation, and test splits, revealing converging performance and reduced variance as confounding intensifies. We validate our method on two clinically relevant tasks, diabetic retinopathy detection from retinal fundus images and tumor detection from brain MRI slices. Then, we further demonstrate its practical utility on another dataset and image modality with a stroke reperfusion prediction task with suspected hidden confounders. This work provides a practical, data-driven diagnostic tool to flag potential confounding and support the reliability assessment of machine learning models in medical imaging. Full article

Heart failure (HF) remains a major global health challenge, necessitating accurate yet accessible diagnostic tools. While the left ventricular ejection fraction (LVEF) is the primary metric for classifying HF into preserved (HFpEF), mid-range (HFmrEF), and reduced (HFrEF) phenotypes, conventional imaging modalities such as echocardiography are resource intensive. In contrast, the electrocardiogram (ECG) offers a low-cost, non-invasive alternative for continuous cardiac assessment. This paper proposes a multi-algorithm artificial intelligence (AI) framework for automated HF phenotype classification using high-resolution ECG signals from 303 patients with chronic heart failure from the MUSIC cohort. After preprocessing (normalization, bandpass filtering), we employed a hybrid approach combining the Pan–Tompkins algorithm for robust R-peak detection with the NeuroKit2 toolbox for the precise delineation of P, Q, S, and T waves. ECG recordings were then segmented using an adaptive beat-centric windowing strategy. From the segmented beats, we extracted a comprehensive set of temporal, morphological, and energy-based features, including RR, QRS, and QT intervals, along with P-wave, QRS-complex, and T-wave energies. These features were used to train and evaluate several ensemble machine learning models—Random Forest, XGBoost, CatBoost, LightGBM, and a stacking classifier—using a stratified 70–15–15 train–validation–test split with 5-fold cross-validation. The LightGBM model achieved the highest performance with a test accuracy of 98.45%, an AUC of 0.9989, and a macro F1-score of 0.9804, outperforming other ensembles and the stacking classifier. The results demonstrate that an AI-driven analysis of ECG-derived morpho-energy features can serve as a reliable, non-invasive screening tool for the accurate and early discrimination of HF phenotypes, potentially supporting clinical decision making and improving patient management in resource-limited settings. Full article

Early and reliable detection of plant diseases is critical for sustaining agricultural productivity and reducing economic losses. In olive cultivation, peacock eye disease poses a significant threat by adversely affecting leaf health and crop yield. While deep learning models have demonstrated strong performance in plant disease detection, their reliance on high-dimensional feature representations often leads to increased computational cost and limited deployability in real-world agricultural settings. This study proposes an efficient and robust olive leaf disease classification framework that integrates deep feature extraction, devised composite filter-based feature selection, and ensemble learning. Deep features are extracted from olive leaf images using transfer learning with ResNet101 and MobileNet architectures. To address feature redundancy and computational inefficiency, multiple filter-based selection strategies—including mutual information, Chi-square, F-score, and five devised composite selectors (score fusion, union, intersection, hybrid, and class-wise filtering)—are employed to generate compact and informative feature subsets of fixed sizes (32, 64, and 128 features). The selected features are evaluated using k-NN, SVM, and LightGBM classifiers under stratified 5-fold cross-validation. Experimental results demonstrate that competitive and near-baseline performance can be achieved with substantially reduced feature dimensionality. In particular, using only 128 selected features, the proposed approach attains up to 0.988 accuracy and 0.976 MCC, closely matching the performance obtained with full deep feature vectors. Furthermore, voting-based ensemble strategies, including iterative majority voting and hybrid GA–BO fusion, further enhance robustness, achieving the highest mean accuracy of 0.9916 among the evaluated ensemble configurations. These findings highlight the effectiveness of the proposed composite filter-based selection and ensemble framework as a practical, lightweight, and accurate solution for olive leaf disease detection, suitable for deployment in precision agriculture and resource-constrained environments. Full article

Background/Objectives: Modern artificial intelligence methods are increasingly used in medical and dental image analysis to support diagnosis and treatment planning. In orthodontics, automatic detection of cephalometric landmarks from X-ray images remains a challenging and clinically relevant task. Methods: This study proposes a multi-model approach for cephalometric landmark detection based on the ALD algorithm and three derived models trained with extended image augmentation techniques. The applied augmentations, including contrast and negative transformations, improved the detection of specific anatomical landmarks. The final detection strategy integrates outputs from all four models, selecting the most accurate prediction for each landmark based on historical performance results. Results: The proposed method was evaluated on real datasets. It achieved a mean radial error (MRE) of 2.12 mm compared to 2.26 mm for the baseline model, and a successful detection rate (SDR) of 72.22% within a 2.5 mm threshold compared to 68.87% for the baseline model. Conclusions: The results demonstrate that the ensemble-based approach improves landmark detection accuracy and has the potential to support clinical orthodontic workflows. Full article

Accurate and automatic segmentation of the liver and liver tumors from computed tomography (CT) images is essential for computer-assisted diagnosis, treatment planning, and clinical decision-making. Although deep learning-based segmentation models, particularly U-Net and its variants, have achieved promising results in medical image analysis, many existing approaches mainly focus on local pixel-level feature extraction and may have limited ability to explicitly model long-range spatial relationships among anatomically meaningful regions. In addition, liver tumor segmentation remains challenging due to low contrast, irregular tumor boundaries, heterogeneous tumor appearances, and noise or artifacts in CT images. To address these limitations, this study proposes a hybrid ensemble neural network architecture that integrates an improved U-Net and a Graph U-Net for automatic liver and liver tumor segmentation. The improved U-Net is designed to capture fine-grained local features and preserve detailed spatial information through an encoder–decoder structure with skip connections, while the Graph U-Net uses Simple Linear Iterative Clustering (SLIC)-based superpixels to construct a graph representation of CT images and model spatial dependencies between adjacent image regions. By combining these complementary representations through an ensemble learning strategy, the proposed framework enhances both pixel-level segmentation accuracy and robustness against noisy imaging conditions. The proposed method was evaluated on the LiTS17 dataset, where CT images were preprocessed using intensity filtering, resizing, data augmentation, and normalization. Experimental results demonstrate that the proposed ensemble architecture achieves 99.2% accuracy for liver segmentation and 98.1% accuracy for liver tumor segmentation, outperforming representative segmentation models such as MultiresUnet and R2U-Net. Furthermore, robustness experiments under different signal-to-noise ratio conditions show that the proposed model maintains stable performance in noisy CT images, achieving 85% accuracy even under severe noise at −4 dB SNR. This result highlights the advantage of integrating convolutional feature learning with graph-based spatial relationship modeling for improving segmentation stability when image quality is degraded by noise or artifacts. These findings indicate that the integration of improved U-Net, SLIC-based graph construction, and Graph U-Net provides an effective and noise-robust solution for liver and liver tumor segmentation, with potential applicability as a computer-assisted tool in clinical image analysis after further validation on larger and external datasets. Full article

Point-of-interest (POI) data are widely used to spatialize and predict socioeconomic variables, yet their consistency across regions and over time, as well as their cross-regional generalizability, remain insufficiently understood. This study examines these issues using county-level units in China for 2010 and 2020 from three perspectives: relationship structure, cross-regional generalization, and model improvement. First, a power-law model is applied to characterize the nonlinear relationship between POI density and population density and to assess its spatiotemporal heterogeneity. Second, generalizability is evaluated by comparing model parameters and predictive performance under random and spatially stratified sampling. Third, multi-source geospatial data, including nighttime lights, road networks, and land use, are integrated to compare linear, spatial, machine learning, and ensemble models. Results reveal a consistent sublinear relationship with strong spatial heterogeneity. Under spatially independent validation, predictive accuracy declines and becomes more variable, indicating limited cross-regional generalization. Integrating multi-source data with ensemble learning improves stability and reduces uncertainty. POI remains the dominant predictor, though its relative importance becomes more concentrated in 2020. Overall, the study highlights the limitations of POI-based population estimation and proposes strategies to enhance robustness and generalizability. Full article

The construction sector is undergoing rapid digital transformation, creating opportunities to enhance environmental safety in urban areas. One critical application lies in air pollution forecasting, particularly regarding fine dust (PM₁₀) emissions. While machine learning (ML) models are widely used for city-wide air quality monitoring, a significant research gap exists in the high-resolution (5 min interval) forecasting of dust at localized “point-pattern” development sites. These densely built urban zones present unique challenges due to highly volatile microclimates and intermittent emission sources that directly affect nearby residents. The purpose of this study is to perform a preliminary performance analysis of eight predictive algorithms—ARIMA, EMA, Prophet, NNAR, Random Forest, SVM, and XGBoost—to identify the most robust approach for short-term PM₁₀ forecasting under limited data (N = 1728). Special attention is paid to the non-linear relationship between meteorological conditions and dust concentrations. Unlike previous studies which focused on general urban backgrounds, this work contributes a validated methodological framework for localized monitoring. The results demonstrate that tree-based ensemble models provide the highest stability and accuracy, offering a reliable basis for future real-time environmental management and active pollution mitigation strategies on urban construction sites. Full article

Soil salinization threatens agricultural sustainability in arid zones, yet quantitative attribution of its spatiotemporal dynamics to multi-source drivers remains scarce at regional scales. To address this, we developed an explainable framework merging Sentinel-1/2, ERA5-Land, and topographic-hydrological indices with XGBoost, trained under weak supervision with proxy labels and independently validated using field-measured ECe. A 7-group, 44-feature ensemble with spatial block 5-fold cross-validation ensured robust assessment. SHapley Additive exPlanations (SHAP) quantified driver contributions and enabled a novel dominant driver zoning (DDZ) framework. Monitoring the Hexi Corridor and Tarim Basin (2017–2024) revealed contrasting trajectories: Hexi’s dynamics were primarily climate-driven (Aridity Index), whereas 19.2% of Tarim showed significant salinization along oasis–desert margins co-dominated by elevation, soil indices, and temperature. The model achieved spatial cross-validation R² values around 0.65. DDZ mapping showed climate dominance in 98.2% of Hexi compared to 76.5% in Tarim, where terrain and optical factors were more influential. The weak supervision strategy overcomes scarce in-situ measurements, while the DDZ maps identified that Land-use-dominated zones recorded the highest salinity, offering clear directives for targeted salinity control in arid basins. Full article

Radiomics enables quantitative medical image analysis by converting imaging data into structured, high-dimensional feature representations for predictive modeling. Despite methodological developments and encouraging retrospective results, radiomics continue to face persistent challenges related to feature instability, limited reproducibility, validation bias, and restricted clinical translation. Existing reviews largely focus on application-specific outcomes or isolated pipeline components, with limited analysis of how interdependent design choices across acquisition, preprocessing, feature engineering, modeling, and evaluation collectively affect robustness and generalizability. This survey provides an end-to-end analysis of radiomics pipelines, examining how methodological decisions at each stage influence feature stability, model reliability, and translational validity. This paper reviews radiomic feature extraction, selection, and dimensionality reduction strategies; classical machine and deep learning–based modeling approaches; and ensemble and hybrid frameworks, with emphasis on validation protocols, data leakage prevention, and statistical reliability. Clinical applications are discussed with a focus on evaluation rigor rather than reported performance metrics. The survey identifies open challenges in standardization, domain shift, and clinical deployment, and outlines future directions such as hybrid radiomics–artificial intelligence models, multimodal fusion, federated learning, and standardized benchmarking. Full article

Accurate carbon price forecasting is critical for trading decisions, risk management and policy formulation in carbon markets. However, mainstream decomposition-ensemble models suffer from two key drawbacks: point-wise modeling fails to capture long-term temporal dependencies, while independent modeling of decomposed trend and seasonal components leads to serious information loss. To address these limitations, this paper proposes a novel Dual-Path Interactive Attention Network (DPIANet) for carbon price time series forecasting, whose dual-parallel architecture consists of a Dual Interaction Attention (DIA) Block and a Decomposition–Subsequence Interaction Attention (DSIA) Block. First, DPIANet employs a patch-wise partitioning strategy to extract local temporal semantic information inaccessible to traditional point-wise segmentation. The DIA Block jointly captures temporal dependencies between different patches within the same sequence and inter-feature dependencies within the same time step. In parallel, the DSIA Block extracts interactive features between decomposed trend and seasonal subsequences, fusing these features with original subsequences to enhance representation and mitigate decomposition-induced information loss. A dual-layer feature selection method (PMI and XGBoost-SHAP) is adopted to identify key driving factors. Experiments on four representative Chinese regional carbon trading markets covering 2014-2020 show that DPIANet achieves superior prediction performance over state-of-the-art models in terms of MSE and MAE, with competitive robustness across different market characteristics, providing practical decision support for carbon market stakeholders. Full article

Optimizing fertilizer management is crucial for increasing crop yields while reducing environmental impact. However, traditional methods rely on extensive field trials, which are costly and limit their scalability. To overcome these limitations, this study developed data-driven yield prediction models (YPM) for wheat, rice, and maize by integrating multiple feature selection and machine learning algorithms with explainable ensemble learning, namely stacking regression (SR) and voting mean (VM). The optimal YPM was subsequently combined with the hybrid optimization strategy to construct an intelligent fertilization decision model (IFDM), and the economic–environmental benefits were subsequently evaluated. The best-performing models were SHAP-SR for wheat and rice and GBM-SR for maize, achieving R² values of 0.79, 0.69, and 0.67, and RMSEs of 681.69, 725.35, and 1091.49 kg ha⁻¹, respectively. Based on the IFDM, the recommended application ranges for nitrogen (N), phosphorus (P₂O₅), and potassium (K₂O) were as follows: for wheat, 122.1–256.3, 45.4–98.2, and 30.6–60.7 kg ha⁻¹; for rice, 170.8–261.2, 55.1–91.4, and 40.6–98.5 kg ha⁻¹; and for maize, 157.5–293.4, 84.2–156.4, and 30.1–62.7 kg ha⁻¹. Simulation-based evaluation suggested that adopting these recommendations could potentially increase average yields by 9.2–12.4% and enhance economic–environmental benefits by 32.86–97.73% across the three crops. This study indicates that coupling interpretable ensemble learning with a hybrid optimization strategy can support efficient decision-making for field-scale fertilization and provides a data-driven and cost-effective approach for precision fertilization, with potential applicability to arid agricultural regions under similar agro-ecological conditions. Full article

Climate change profoundly impacts insect distribution and ecological functions. For the predatory robber flies Microstylum dux and M. oberthurii (Diptera: Asilidae), clarifying their distribution and climatic responses is vital for natural enemy conservation and biological control. Using a parameter-optimized biomod2 ensemble model, we predicted their potential distributions under current and future climates, and analyzed key variables, centroid shifts, and niche dynamics. Current suitable habitats concentrate in southeast China and are scarce in the northwest. Future total suitable area remains stable but structurally reorganizes, with highly suitable habitats expanding and moderately suitable ones contracting. Key drivers are precipitation of the coldest quarter (bio19) and mean diurnal range (bio2). Habitat centroids migrate westward or southwestward with fluctuating range expansion. M. dux is a niche specialist (niche width = 0.257), while M. oberthurii is a generalist (niche width = 0.539). Their niche overlap shows non-linear “divergence-convergence-divergence” dynamics. This study supports natural enemy conservation and biological control strategy formulation. Full article