High-dimensional gene expression data pose substantial challenges for machine-learning-based diagnostic modelling due to extreme dimensionality, noise, heterogeneous measurement conditions, and limited sample sizes. These factors strongly affect classifier behaviour and necessitate systematic comparative evaluation under controlled experimental conditions. This study presents a comprehensive assessment of widely used machine-learning classifiers applied to high-dimensional transcriptomic data, with a focus on robustness, stability, and suitability for gene expression-driven decision-support systems. The analysis covers representative model families, including tree-based and ensemble methods, margin-based classifiers, and regularized linear models. All classifiers were evaluated within a unified analytical pipeline incorporating normalization, variance-based filtering, stratified data partitioning, Bayesian hyperparameter optimization, and repeated validation. Performance was assessed using accuracy, precision, recall, F1-score, and balanced accuracy to account for class imbalance and heterogeneous sample structures. Experimental results reveal pronounced differences in classifier performance under high-dimensional conditions. Ensemble-based approaches, particularly Random Forest and gradient boosting models, consistently achieved higher stability and balanced accuracy across heterogeneous datasets. Regularized linear models with elastic-net penalties performed competitively after dimensionality reduction, while support vector machines demonstrated high accuracy but required careful kernel and regularization tuning. In contrast, probabilistic classifiers exhibited increased sensitivity to noise and distributional heterogeneity. Overall, the findings emphasize the importance of ensemble learning and regularization strategies for reliable classification of transcriptomic data and provide a reproducible methodological framework for information-technology-driven diagnostic modelling based on gene expression profiles.
Financial systems are highly nonlinear and often influenced by investment cycles, market fluctuations, and external financing activities, which can generate complex dynamic behaviors. This paper proposes a new three-dimensional chaotic financial firm model by extending the classical Bouali financial system through the inclusion of a cyclic reinvestment perturbation represented by a sinusoidal nonlinear term. The proposed modification aims to capture state-dependent nonlinear reinvestment feedback in reinvestment decisions caused by changing economic conditions and business cycles. The dynamical characteristics of the model are investigated using equilibrium analysis, local stability theory, Lyapunov exponents, the Kaplan–Yorke dimension, bifurcation analysis, and multistability analysis. Numerical results show that the system possesses three equilibrium points, all of which are unstable under the selected parameter setting. The system exhibits chaotic dynamics confirmed by a positive maximum Lyapunov exponent and a fractal attractor characterized by a Kaplan–Yorke dimension greater than two. Comparative results indicate that the new model demonstrates richer nonlinear dynamical behavior than existing financial firm systems reported in the literature. Furthermore, bifurcation and Lyapunov spectrum analyses disclose multiple transitions between periodic and chaotic states as system parameters vary, while multistability analysis reveals the coexistence of symmetric periodic and chaotic attractors under identical parameter values but different initial conditions. Finally, total amplitude control and offset boosting techniques are implemented to regulate attractor size and position while preserving the underlying chaotic behavior. These findings demonstrate that cyclic reinvestment perturbations significantly enrich the nonlinear dynamics and symmetry properties of financial firm systems, providing a useful framework for studying complex financial behaviors and chaos control.
Cryopathy syndromes are difficult to classify because laboratory patterns often overlap across diagnostic categories, while some diagnoses are rare. This makes routine interpretation of cryoglobulin-related tests challenging and increases dependence on expert judgment. The aim of this study was to develop and compare machine learning approaches for automated classification of cryopathy syndromes from laboratory data and to identify a practical strategy for clinical decision support. Methods: We analysed laboratory records from 2,686 patients assigned to 14 diagnostic categories. The dataset included demographic variables, cryoglobulin measurements, precipitation tests, and hemagglutinin and hemolysin titers. Data preprocessing included cleaning, encoding, imputation, normalization, and construction of clinically informed interaction features. We evaluated 12 modelling strategies, including Random Forest, Gradient Boosted Trees, Multi-Layer Perceptron, soft-voting ensembles, class balancing with Synthetic Minority Over-sampling Technique, hierarchical classification, period-aware models, targeted binary classifiers, and probability calibration. Performance was assessed using stratified train-test evaluation and stratified 5-fold cross-validation. The main metrics were macro-averaged F1 score, accuracy, Top-3 accuracy, and expected calibration error. The overall task proved difficult because of marked class imbalance and clinical overlap between diagnoses. The best multiclass performance was achieved by a soft-voting ensemble of Random Forest and Gradient Boosted Trees. Cross-validation confirmed stable performance for the balanced Random Forest model. Tree-based methods consistently outperformed the neural network model. Feature engineering improved discrimination, and the most informative predictors were derived cryoglobulin-based interaction features.
Background. Fusion of polarization images with conventional thermal infrared images combines complementary physical information to improve scene interpretation. Infrared intensity imaging captures thermal emission and emissivity patterns. In contrast, polarization parameters, including the degree and angle of linear polarization, encode surface microstructure, material composition, and geometric orientation that are inaccessible to intensity-only sensing. Their fusion enhances spatial detail, target discrimination, and robustness under degraded visibility conditions, reflecting a broader shift toward multimodal perception. Materials and Methods. This study provides an analytical review of polarization–infrared fusion techniques. The development of methods is examined from classical linear transforms, multiresolution decomposition, and energy-based decision rules to adaptive hybrid frameworks. Particular attention is given to deep learning architectures and physics-informed models that incorporate polarization formation principles and radiometric constraints. The review situates fusion progress within advances in sensor technologies and computational imaging. Results and Discussion. The analysis indicates that the incorporation of polarization information substantially enhances fused image quality compared with conventional intensity-based techniques. Classical algorithms provide stable and interpretable performance but exhibit limited adaptability to complex environments. Learning-based and hybrid strategies demonstrate superior contrast preservation, structural detail enhancement, and target detectability, especially in thermally heterogeneous or visually degraded scenarios. Persistent challenges include limited annotated datasets, polarization-induced noise sensitivity, and reduced physical interpretability of deep neural representations. Conclusion. Polarization–infrared fusion represents a transition from mono-modality to multi-modal observation with the system-oriented perception integrating physical modeling with data-driven learning. Future research should emphasize physically reasoned neural design, standardized evaluation protocols, and improved interpretability to support reliable real-world deployment.
Context: Green infrastructure (GI) is an EU policy concept addressing functional connectivity to support biodiversity conservation and ecosystem services. In regions with significant forest fragmentation, establishing functional GI presents a challenge. Objectives We evaluate the contribution of low-productive forests (LPF) in networks of existing High Conservation Value Forests (HCVF) in Sweden, and discuss conservation implications for GI across different forest types and ecoregions. Methods Three characteristics for LPF were assessed: area extent of different forest types, patch size distribution, and role in functional landscape connectivity across five ecoregions. The primary data sources were a database of known HCVF and the national land cover data that includes data on LPF. Results LPF are dominant in the mountain region, both for all forests and among HCVF classified as low-productive. For other ecoregions, the extent of LPF was lower. Pine forests dominate LPF in all regions below the mountain region. For pine forests outside the mountain region, the inclusion of LPF increased the area of functional HCVF habitat for forest generalist species by 5–8% and for forest specialists by 6–16%. For other forest types, there was minimal improvement in connectivity. Overall, most LPF occur as small fragments, particularly when considering individual forest types. Conclusion LPF can strengthen GI in some regions and some forest types. However, they cannot replace the need for additional protection and restoration of productive forests to ensure functional connectivity. The role of LPFs as a contribution to conservation objectives is limited and related to species confined to LPFs.
