1. Joint modeling of grain yield and root lodging in maize using multi-output neural network and machine learning models under defined environmental conditionsDušan Dunđerski, Božana Purar, Anja Đurić, Maja Tanasković, Dušan Stanisavljević, 2026, izvirni znanstveni članek Povzetek: We evaluated a multi-output neural network framework for jointly analyzing maize grain yield (GY) and root lodging percentage (LP) using above-ground morphological traits measured under defined environmental conditions. To address model robustness, the multi-output neural network was compared with linear regression, elastic net, random forest, and XGBoost using repeated five-fold cross-validation, an 80/20 holdout split, and independent year-wise validation. Under repeated cross-validation, XGBoost provided the strongest average predictive performance for both traits, with R2 values of 0.57 for GY and 0.67 for LP. The multi-output neural network showed moderate performance, with R2 values of 0.49 for GY and 0.57 for LP. Final holdout performance for the neural network for GY and LP was R2 = 0.64 and R2 = 0.92, respectively. Year-wise validation showed weak temporal transferability because the two seasons differed not only in environmental conditions, but also in lodging mechanism. Repeated permutation importance identified ear width (EW), kernel row number (RNE), thousand kernel mass (KM1000), and kernel number per ear (KNE) as important predictors of GY, while LP prediction was most strongly associated with internode major diameter (IDmajor), ear length (EL), and the number of green leaves (NGL). Across both permutation importance and SHAP, only RNE and NGL were consistently shared between GY and LP. Supplementary ALE diagnostics indicated that RNE showed increasing model-estimated effects for both predicted GY and LP, whereas NGL showed a positive association with predicted GY but a decreasing or nonlinear association with predicted LP. These results show that joint modeling can support exploratory trait interpretation, but the predictive relationships remain environment-specific and should not be interpreted as causal or broadly transferable without further multi-environment validation.
Ključne besede: machine learning, grain yield, lodging, maize, permutation feature importance, joint modeling, environment
Objavljeno v DiRROS: 23.06.2026; Ogledov: 153; Prenosov: 102
 Celotno besedilo (862,14 KB)
Gradivo ima več datotek! Več... |
2. An adaptive graph-based method for structured learning and decisiona nalysisSovan Samanta, Tofigh Allahviranloo, Leo Mršić, Antonios Kalampakas, 2026, izvirni znanstveni članek Ključne besede: machine learning, knowledge engineering, quantum graph structure, parameter space, federated learning, analytical modeling
Objavljeno v DiRROS: 17.06.2026; Ogledov: 118; Prenosov: 122
Celotno besedilo (861,68 KB)
Gradivo ima več datotek! Več... |
3. Video enhancement for increased spatio-temporal resolution in thermal videos : demonstration on a pool fireAndrea Lucherini, Steven Verstockt, Bart Merci, 2026, izvirni znanstveni članek Povzetek: A spatio-temporal video enhancement of a small-scale pool fire is performed to address the typically low spatial resolution and frame rate of inexpensive infrared (IR) cameras. Improving image quality can increase the applicability of low-cost thermal cameras for certain research tasks and analyses. The spatial resolution and frame rate are doubled, from 310 × 250 pixels (px) to 620 × 500 px, and from 25 frames per second (fps) to 50 fps, as well as from 50 fps to 100 fps. Spatial resolution enhancement is achieved using super-resolution methods based on deep learning, employing several pre-trained models: Fast Super-Resolution CNN (FSRCNN), Efficient Sub-Pixel Convolutional Network (ESPCN), Enhanced Deep Super-Resolution (EDSR), Laplacian Pyramid Super-Resolution Network (LapSRN), and Real-ESRGAN. The footage consists of an n-heptane pool fire recorded using a mid-wave infrared (MWIR) FLIR X6981 HS InSb camera. EDSR provides the best performance for both purely resized images and images subjected to complex degradation. For temporal enhancement, a pre-trained frame interpolation model, FLAVR (FlowAgnostic Video Representation), is used. The resulting interpolated frames appear realistic and preserve the overall flow direction and shape of the flame. The interpolated frames are compared with ground-truth data to validate the accuracy of the temporal enhancement.
Ključne besede: image processing, thermal camera, machine learning, pool fire
Objavljeno v DiRROS: 15.06.2026; Ogledov: 167; Prenosov: 175
Celotno besedilo (6,85 MB)
Gradivo ima več datotek! Več... |
4. Evaluation of deep learning models for image-based classification of timber logs by market valueMatevž Triplat, Žiga Lukančič, Vasja Kavčič, 2026, izvirni znanstveni članek Povzetek: The identification of standing tree species, timber logs, and on-site assessment of their quality and value using images holds significant potential for forestry applications, including inventory management, traceability under EU regulations like the Deforestation Regulation, and market valuation amid growing demands for sustainable practices. This study addresses this by classifying images of timber logs by tree species and market value using the Orange data mining software, which leverages pre-trained convolutional neural networks (Inception v3 and SqueezeNet) to generate embeddings from a dataset of 5549 images collected at a real timber auction in Slovenia, followed by logistic regression image classification. Results show high accuracy for tree species classification (up to 92.6%), but substantially lower accuracy for market value classification (40%–55%), reflecting the greater complexity of value determination from visual features. These findings underscore the promise of deep learning for species identification while indicating the need for further methodological advancements to enhance value classification reliability, which offers the practical impact for operational forestry and bioeconomy value chains.
Ključne besede: image classification, timber quality, high value assortments, auctions, wood products, convolutional neural networks, CNNs, non-destructive evaluation, machine learning in forestry, tree species image recognition, forest wood assortment value
Objavljeno v DiRROS: 12.06.2026; Ogledov: 182; Prenosov: 155
Celotno besedilo (4,84 MB)
Gradivo ima več datotek! Več... |
5. Differentiating viral and bacterial infections : a machine learning model based on routine blood test valuesGregor Gunčar, Matjaž Kukar, Tim Smole, Sašo Moškon, Tomaž Vovko, Simon Podnar, Peter Černelč, Miran Brvar, Mateja Notar, Manca Köster, Marjeta Tušek Jelenc, Žiga Osterc, Marko Notar, 2024, izvirni znanstveni članek Ključne besede: viruses, bacteria, machine learning
Objavljeno v DiRROS: 09.06.2026; Ogledov: 122; Prenosov: 188
Celotno besedilo (3,89 MB)
Gradivo ima več datotek! Več... |
6. Wearable online freezing of gait detection and cueing systemJan Slemenšek, Jelka Geršak, Božidar Bratina, Vesna M. Van Midden, Zvezdan Pirtošek, Riko Šafarič, 2024, izvirni znanstveni članek Povzetek: This paper presents a real-time wearable system designed to assist Parkinson’s disease patients experiencing freezing of gait episodes. The system utilizes advanced machine learning models, including convolutional and recurrent neural networks, enhanced with past sample data preprocessing to achieve high accuracy, efficiency, and robustness. By continuously monitoring gait patterns, the system provides timely interventions, improving mobility and reducing the impact of freezing episodes. This paper explores the implementation of a CNN+RNN+PS machine learning model on a microcontroller-based device. The device operates at a real-time processing rate of 40 Hz and is deployed in practical settings to provide ‘on demand’ vibratory stimulation to patients. This paper examines the system’s ability to operate with minimal latency, achieving an average detection delay of just 261 milliseconds and a freezing of gait detection accuracy of 95.1%. While patients received on-demand stimulation, the system’s effectiveness was assessed by decreasing the average duration of freezing of gait episodes by 45%. These preliminarily results underscore the potential of personalized, real-time feedback systems in enhancing the quality of life and rehabilitation outcomes for patients with movement disorders.
Ključne besede: Parkinson’s disease, freezing of gait, machine learning, real-time systems, wearable devices, on-demand stimulation
Objavljeno v DiRROS: 04.06.2026; Ogledov: 182; Prenosov: 128
Celotno besedilo (6,28 MB)
Gradivo ima več datotek! Več... |
7. Deregulation in adult IgA vasculitis skin as the basis for the discovery of novel serum biomarkersMatija Bajželj, Matjaž Hladnik, Rok Blagus, Vesna Jurčić, Ana Markež, Tanya Deniz Toluay, Snežna Sodin-Šemrl, Alojzija Hočevar, Katja Lakota, 2024, izvirni znanstveni članek Povzetek: Introduction: Immunoglobulin A vasculitis (IgAV) in adults has a variable disease course, with patients often developing gastrointestinal and renal involvement and thus contributing to higher mortality. Due to understudied molecular mechanisms in IgAV currently used biomarkers for IgAV visceral involvement are largely lacking. Our aim was to search for potential serum biomarkers based on the skin transcriptomic signature. Methods: RNA sequencing analysis was conducted on skin biopsies collected from 6 treatment-naïve patients (3 skin only and 3 renal involvement) and 3 healthy controls (HC) to get insight into deregulated processes at the transcriptomic level. 15 analytes were selected and measured based on the transcriptome analysis (adiponectin, lipopolysaccharide binding protein (LBP), matrix metalloproteinase-1 (MMP1), C-C motif chemokine ligand (CCL) 19, kallikrein-5, CCL3, leptin, C-X-C motif chemokine ligand (CXCL) 5, osteopontin, interleukin (IL)-15, CXCL10, angiopoietin-like 4 (ANGPTL4), SERPIN A12/vaspin, IL-18 and fatty acid-binding protein 4 (FABP4)) in sera of 59 IgAV and 22 HC. Machine learning was used to assess the ability of the analytes to predict IgAV and its organ involvement. Results: Based on the gene expression levels in the skin, we were able to differentiate between IgAV patients and HC using principal component analysis (PCA) and a sample-to-sample distance matrix. Differential expression analysis revealed 49 differentially expressed genes (DEGs) in all IgAV patient's vs. HC. Patients with renal involvement had more DEGs than patients with skin involvement only (507 vs. 46 DEGs) as compared to HC, suggesting different skin signatures. Major dysregulated processes in patients with renal involvement were lipid metabolism, acute inflammatory response, and extracellular matrix (ECM)-related processes. 11 of 15 analytes selected based on affected processes in IgAV skin (osteopontin, LBP, ANGPTL4, IL-15, FABP4, CCL19, kallikrein-5, CCL3, leptin, IL-18 and MMP1) were significantly higher (p-adj < 0.05) in IgAV serum as compared to HC. Prediction models utilizing measured analytes showed high potential for predicting adult IgAV. Conclusion: Skin transcriptomic data revealed deregulations in lipid metabolism and acute inflammatory response, reflected also in serum analyte measurements. LBP, among others, could serve as a potential biomarker of renal complications, while adiponectin and CXCL10 could indicate gastrointestinal involvement.
Ključne besede: acute inflammatory response, adults, IgA vasculitis, lipid metabolism, machine learning, RNA sequencing, serum biomarkers
Objavljeno v DiRROS: 04.06.2026; Ogledov: 171; Prenosov: 120
Celotno besedilo (1,70 MB)
Gradivo ima več datotek! Več... |
8. Machine learning-driven mapping of prokaryotic community diversity in the Mediterranean Sea using omics, earth observation, and model dataChristian Marchese, Maria Laura Zoffoli, Pierre Ramond, Tinkara Tinta, Neža Orel, 2026, izvirni znanstveni članek Povzetek: Marine prokaryotic communities are major contributors to oceanic food webs and global biogeochemical cycles. However, basin-scale diversity patterns and environmental drivers remain poorly understood. In this study, we applied a machine-learning framework to model the diversity of marine prokaryotic communities across the Mediterranean Sea. Diversity was quantified using the Shannon Diversity Index (SDI) derived from 16S rRNA gene sequencing. The in situ dataset included ~600 samples collected year-round from 2001 to 2023 at coastal and open-water sites, providing broad temporal coverage and multisite spatial sampling. We trained an XGBoost model using satellite-derived and modeled oceanographic variables matched to the SDI observations. The model achieved robust predictive performance (R2 = 0.78 for training and 0.70 for testing, with RMSE = 0.31 and MAPE = 0.05 across both) and captured broad basin spatial and seasonal patterns in prokaryotic community diversity, with greater uncertainty in less-represented regions. Diversity was highest in nutrient-rich coastal areas and during winter mixing, and lowest in summer-stratified or oligotrophic waters. SHAP analysis identified photoperiod as the most significant predictor, underscoring the central role of seasonal light cycles in shaping prokaryotic community diversity. Other predictors exhibited significant season- and region-dependent effects, each contributing positively within specific environmental thresholds. Climatological diversity maps revealed consistent spatiotemporal patterns, highlighting a notable west-to-east decrease in diversity and coastal hotspots. These results demonstrate that machine learning can identify major environmental drivers of prokaryotic diversity and upscale discrete observations to basin-wide predictions. This approach is transferable to other planktonic groups and supports scalable ecosystem monitoring across environmental gradients.
Ključne besede: prokaryotic community diversity, machine learning, Mediterranean Sea, omics, Shannon diversity index, remote sensing
Objavljeno v DiRROS: 04.05.2026; Ogledov: 252; Prenosov: 296
Celotno besedilo (11,44 MB)
Gradivo ima več datotek! Več... |
9. Ethical considerations on the use of big data and artificial intelligence in kidney research from the ERA ethics committeeWim Van Biesen, Jadranka Buturović-Ponikvar, Monica Fontana, Peter Heering, Mehmet S. Sever, Simon Sawhney, Valerie Luyckx, 2025, pregledni znanstveni članek Povzetek: In the current paper, we will focus on requirements to ensure big data can advance the outcomes of our patients suffering from kidney disease. The associated ethical question is whether and how we as a nephrology community can and should encourage the collection of big data of our patients. We identify some ethical reflections on the use of big data, and their importance and relevance. Furthermore, we balance advantages and pitfalls and discuss requirements to make legitimate and ethical use of big data possible. The collection, organization, and curation of data come upfront in the pipeline before any analyses. Great care must therefore be taken to ensure quality of the data at this stage, to avoid the ‘garbage in garbage out’ problem and suboptimal patient care as a consequence of such analyses. Access to the data should be organized so that correct and efficient use of data is possible. This means that data must be stored safely, so that only those entitled to do so can access them. At the same time, those who are entitled to access the data should be able to do so in an efficient way, so as not to hinder relevant research. Analysis of observational data is itself prone to many errors and biases. Each of these biases can finally result in provision of low-quality medical care. Secure platforms should therefore also ensure correct methodology is used to interpret the available data. This requires close collaboration of a skilled workforce of experts in medical research and data scientists. Only then will our patients be able to benefit fully from the potential of AI and big data.
Ključne besede: artificial intelligence, big data, machine learning, observational trial
Objavljeno v DiRROS: 22.04.2026; Ogledov: 183; Prenosov: 180
Celotno besedilo (705,70 KB)
Gradivo ima več datotek! Več... |
10. Validation of reading as a predictor of mild cognitive impairmentVida Groznik, Martin Možina, Timotej Lazar, Dejan Georgiev, Aleš Semeja, Aleksander Sadikov, 2025, izvirni znanstveni članek Ključne besede: eye-tracking, machine learning, mild cognitive impairment, validation, reading characteristics
Objavljeno v DiRROS: 20.04.2026; Ogledov: 197; Prenosov: 223
Celotno besedilo (1,68 MB)
Gradivo ima več datotek! Več... |
