Learning-based Modeling Research Articles

Exploring trends and emerging topics in oceanography (1992–2021) using deep learning-based topic modeling and cluster analysis

Recent advancements in data sampling, modeling, and increased collaboration among scientists have driven a substantial rise in oceanography publications. By utilizing big data and deep learning, we built a BERTopic model to perform topic modeling on 334,765 publications from the Web of Science, spanning 1992 to 2021, to examine research trends and emerging topics in oceanography. We then created a topic cluster map containing the 100 most popular topics, revealing the complex interplay of topics across different sub-fields and the interdisciplinary nature of oceanography. Key emerging topics, such as precipitation and climate, reflect the increasing focus on climate change and underscore the ocean’s vital role in addressing global environmental challenges. We also revealed geographical trends in research focus and collaboration patterns. Our study not only highlights trends and areas of growing interest but also identifies topics requiring further studies, emphasizing the importance of international collaboration for future research.

Deep Learning-Based Eddy Viscosity Modeling for Improved RANS Simulations of Wind Pressures on Bluff Bodies

Deep Learning-Based Eddy Viscosity Modeling for Improved RANS Simulations of Wind Pressures on Bluff Bodies

Deep learning-based predictive modeling of transonic flow over an airfoil

Effectively predicting transonic unsteady flow over an airfoil presents significant challenges due to its complex dynamics. In this study, we utilize a deep neural network architecture designed to capture intricate flow behavior. Through comprehensive training, our models successfully represent the complexities of transonic and unsteady flow, even under previously unseen conditions. By leveraging the differentiable nature of neural network representations, we develop a framework for evaluating fundamental physical properties using linear stability analysis. This approach bridges neural network modeling with traditional modal analysis, providing critical insights into transonic flow dynamics while improving the interpretability of neural network-based flow diagnostics.

Performance prediction of sintered NdFeB magnet using multi-head attention regression models

The preparation of sintered NdFeB magnets is complex, time-consuming, and costly. Data-driven machine learning methods can enhance the efficiency of material synthesis and performance optimization. Traditional machine learning models based on mathematical and statistical principles are effective for structured data and offer high interpretability. However, as the scale and dimensionality of the data increase, the computational complexity of models rises dramatically, making hyperparameter tuning more challenging. By contrast, neural network models possess strong nonlinear modeling capabilities for handling large-scale data, but their decision-making and inferential processes remain opaque. To enhance interpretability of neural network, we collected 1,200 high-quality experimental data points and developed a multi-head attention regression model by integrating an attention mechanism into the neural network. The model enables parallel data processing, accelerates both training and inference speed, and reduces reliance on feature engineering and hyperparameter tuning. The coefficients of determination for remanence and coercivity are 0.97 and 0.84, respectively. This study offers new insights into machine learning-based modeling of structure-property relationships in materials and has potential to advance the research of multimodal NdFeB magnet models.

Machine learning-based modeling of ground level ozone formation in Bangalore and New Delhi cities in India

Machine learning-based modeling of ground level ozone formation in Bangalore and New Delhi cities in India

Determining plasma dose using equivalent total oxidation potential (ETOP): Concept to practical application via machine learning

Atmospheric pressure nonequilibrium plasma holds significant potential in biomedical applications due to its ability to generate reactive species at low temperatures. However, accurately quantifying and controlling plasma dosage remains challenging. Although equivalent total oxidation potential (ETOP) has been proposed for defining dosage, previous methods required measurement of various reactive oxygen and nitrogen species (RONS) densities, which are impractical in diverse plasma settings. Efficient ETOP prediction across variable conditions is thus essential. To address this, we propose a machine learning-based ETOP modeling method. This study collected RONS density data under various conditions using laser-induced fluorescence and trained an artificial neural network to predict ETOP values based on input parameters like voltage, gas flow rate, oxygen concentration, and humidity. This approach enables efficient ETOP prediction across variable conditions, supporting the standardization and clinical application of plasma medicine.

Combating Antimicrobial Resistance Through a Data-Driven Approach to Optimize Antibiotic Use and Improve Patient Outcomes: Protocol for a Mixed Methods Study.

It is projected that drug-resistant infections will lead to 10 million deaths annually by 2050 if left unabated. Despite this threat, surveillance data from resource-limited settings are scarce and often lack antimicrobial resistance (AMR)-related clinical outcomes and economic burden. We aim to build an AMR and antimicrobial use (AMU) data warehouse, describe the trends of resistance and antibiotic use, determine the economic burden of AMR in Uganda, and develop a machine learning algorithm to predict AMR-related clinical outcomes. The overall objective of the study is to use data-driven approaches to optimize antibiotic use and combat antimicrobial-resistant infections in Uganda. We aim to (1) build a dynamic AMR and antimicrobial use and consumption (AMUC) data warehouse to support research in AMR and AMUC to inform AMR-related interventions and public health policy, (2) evaluate the trends in AMR and antibiotic use based on annual antibiotic and point prevalence survey data collected at 9 regional referral hospitals over a 5-year period, (3) develop a machine learning model to predict the clinical outcomes of patients with bacterial infectious syndromes due to drug-resistant pathogens, and (4) estimate the annual economic burden of AMR in Uganda using the cost-of-illness approach. We will conduct a study involving data curation, machine learning-based modeling, and cost-of-illness analysis using AMR and AMU data abstracted from procurement, human resources, and clinical records of patients with bacterial infectious syndromes at 9 regional referral hospitals in Uganda collected between 2018 and 2026. We will use data curation procedures, FLAIR (Findable, Linkable, Accessible, Interactable and Repeatable) principles, and role-based access control to build a robust and dynamic AMR and AMU data warehouse. We will also apply machine learning algorithms to model AMR-related clinical outcomes, advanced statistical analysis to study AMR and AMU trends, and cost-of-illness analysis to determine the AMR-related economic burden. The study received funding from the Wellcome Trust through the Centers for Antimicrobial Optimisation Network (CAMO-Net) in April 2023. As of October 28, 2024, we completed data warehouse development, which is now under testing; completed data curation of the historical Fleming Fund surveillance data (2020-2023); and collected retrospective AMR records for 599 patients that contained clinical outcomes and cost-of-illness economic burden data across 9 surveillance sites for objectives 3 and 4, respectively. The data warehouse will promote access to rich and interlinked AMR and AMU data sets to answer AMR program and research questions using a wide evidence base. The AMR-related clinical outcomes model and cost data will facilitate improvement in the clinical management of AMR patients and guide resource allocation to support AMR surveillance and interventions. PRR1-10.2196/58116.

Machine learning-based modeling of heating process, case study: a greenhouse prototype

ABSTRACT Heating demand is the most energy-consuming aspect of controlled environment agriculture (CEA) and seeks the attention of researchers and the commercial side of the industry. Here an experimental methodology to model the microclimate by considering the capacity of the heater as a variable is introduced. Different tests were applied on an exclusive prototype with broad capabilities to prepare sets of data for developing the noted model. A four-layered fully connected neural network which is considered one of the most used machine learning approaches was employed due to its perception for this application. Five different heater capacities with the same initial conditions warmed the microclimate till reaching the steady-state condition. These data were used to train the algorithm capable of predicting the whole process of heating. Two other tests, one with a capacity inside the range and one outside, were conducted additionally in order to validate the model. This model can be used in various optimization problems with great width. It can be used to run different types of simulations to solve an optimization problem in which minimum energy is desired, or to find the optimum heating capacity which satisfies the needs of a specific greenhouse.

Machine learning-based surrogate resilience modeling for preliminary seismic design

Early decisions made during the preliminary seismic design, particularly regarding the stiffness and strength of the selected lateral force-resisting system, have a significant impact on seismic resilience. However, the effects of these crucial decisions on resilience are not explicitly assessed until the end of the routine code-based design. The indirect design approach can bring about substantial design iterations that involve computationally expensive resilience assessment. For the support of direct resilience-informed design, this paper employs machine learning techniques to develop a surrogate model that maps site and building features to resilience indices. The surrogate model aids design practitioners in selecting satisfying design combinations of stiffness and strength in the preliminary seismic design. It provides real-time feedback on the resulting resilience, drastically reducing the time and effort required for iterative design revisions. To demonstrate the effectiveness of the proposed approach, a data set of 10,000 steel moment-resisting frames (SMRFs) with wide-ranging design combinations is created; additionally, six regression algorithms, including linear regression, support vector regression, multi-layer perceptron, k-nearest neighbors, decision tree, and extreme gradient boosting (XGBoost), are used to establish the surrogate model. Among the six algorithms, XGBoost performed the best in terms of the lowest RMSE (0.031) and the highest R2 (0.97) for the test set. Identified by the SHapley Additive exPlanations (SHAP) approach, the seismic hazard-related design basic acceleration and the stiffness-related building period coefficient have the most significant impact on the prediction of XGBoost. This enables a physical and quantitative understanding of the surrogate estimates.

Machine Learning-Based Modeling for Structural Engineering: A Comprehensive Survey and Applications Overview

Modeling and simulation have been extensively used to solve a wide range of problems in structural engineering. However, many simulations require significant computational resources, resulting in exponentially increasing computational time as the spatial and temporal scales of the models increase. This is particularly relevant as the demand for higher fidelity models and simulations increases. Recently, the rapid developments in artificial intelligence technologies, coupled with the wide availability of computational resources and data, have driven the extensive adoption of machine learning techniques to improve the computational accuracy and precision of simulations, which enhances their practicality and potential. In this paper, we present a comprehensive survey of the methodologies and techniques used in this context to solve computationally demanding problems, such as structural system identification, structural design, and prediction applications. Specialized deep neural network algorithms, such as the enhanced probabilistic neural network, have been the subject of numerous articles. However, other machine learning algorithms, including neural dynamic classification and dynamic ensemble learning, have shown significant potential for major advancements in specific applications of structural engineering. Our objective in this paper is to provide a state-of-the-art review of machine learning-based modeling in structural engineering, along with its applications in the following areas: (i) computational mechanics, (ii) structural health monitoring, (iii) structural design and manufacturing, (iv) stress analysis, (v) failure analysis, (vi) material modeling and design, and (vii) optimization problems. We aim to offer a comprehensive overview and provide perspectives on these powerful techniques, which have the potential to become alternatives to conventional modeling methods.

Hydrodynamic Shape Optimization of a Naval Destroyer by Machine Learning Methods

This paper explores the integration of advanced machine learning (ML) techniques within simulation-based design optimization (SBDO) processes for naval applications, focusing on the hydrodynamic shape optimization of the DTMB 5415 destroyer model. The use of unsupervised learning for design-space dimensionality reduction, combined with supervised learning through active learning-based multi-fidelity surrogate modeling, allows for significant improvements in computational efficiency while addressing complex, high-dimensional design spaces. By applying these ML techniques to both single- and multi-objective optimizations, aimed at minimizing resistance and enhancing seakeeping performance, the proposed framework demonstrates its practical value in hydrodynamic design. This approach provides a scalable and efficient solution, reducing the reliance on high-fidelity simulations while accelerating the optimization process, without substantial modifications to existing toolchains. A design-space dimensionality reduction of approximately 70% is achieved, reducing the design variables from 22 to 7 while retaining 95% of the original geometric variance. Additionally, computational cost reductions of 65% to 98% are observed, compared to using the full design space and high-fidelity simulations only.

Machine Learning-Based Path Loss Modeling With Simplified Features

Machine Learning-Based Path Loss Modeling With Simplified Features

Efficient proxy for time-lapse seismic forward modeling using a U-net encoder-decoder approach

The time-lapse seismic (4D seismic) forward modeling provides crucial data for calibrating reservoir models through different data assimilation algorithms. Unfortunately, the traditional 4D seismic forward-modeling methodology is time-expensive and entails significant computational resource consumption. To address these drawbacks, in this work, our goal is to develop a proxy model for the 4D seismic forward modeling using a class of machine learning algorithm named U-Net encoder–decoder. We applied the developed proxy model to a benchmark carbonate reservoir using an ensemble of reservoir simulation models from UNISIM IV dataset (a synthetic benchmark based on real data of a Brazilian pre-salt field). Moreover, we aim to introduce seminal strategies for interpreting the proposed proxy model operation, its outputs, and possible correlations between input and output variables. To achieve this, we trained and tested two versions of U-net-based models and applied methods for explainable artificial intelligence, such as Gradient-weighted Class Activation Mapping (Grad-CAM) and Forward Feature Selection. The experiments showed good results when applied to the test dataset. The correlation coefficient (R2) values were in the range of 0.7 to 0.9, showing the efficiency of the proxy model to replace the 4D seismic forward modeling. Through qualitative analysis, it was possible to identify which input properties and regions of the reservoir are more relevant for the model’s inference. These results are a step toward robust, explainable machine learning-based proxy forward modeling.

An integrated framework of deep learning and entropy theory for enhanced high-dimensional permeability field identification in heterogeneous aquifers

Accurately estimating high-dimensional permeability (k) fields through data assimilation is critical for minimizing uncertainties in groundwater flow and solute transport simulations. However, designing an effective monitoring network to obtain diverse system responses in heterogeneous aquifers for data assimilation presents significant challenges. To investigate the influence of different measurement types (hydraulic heads, solute concentrations, and permeability) and monitoring strategies on the accuracy of permeability characterization, this study integrates a deep learning-based surrogate modeling approach and the entropy-based maximum information minimum redundancy (MIMR) monitoring design criterion into a data assimilation framework. An ensemble MIMR-optimized method is developed to provide more comprehensive monitoring information and avoid missing key information due to the randomness of stochastic response datasets in entropy analysis. A numerical case of solute transport with log-Gaussian permeability fields is presented, with twelve scenarios designed by combining different measurement types and monitoring strategies. The results demonstrated that the proposed ensemble MIMR-optimized method significantly improved the k-field estimates compared to the conventional MIMR method. Additionally, high prediction accuracy in forward modeling is essential for ensuring reliable inversion results, especially for observation data with strong nonlinearity. The findings of this study enhance our understanding and management of k-field estimation in heterogeneous aquifers, contributing to the development of more robust inversion frameworks for general data assimilation tasks.

Group interaction effect on breaking wave forces on a vertical pile: Experimental tests and predictive models

Pile groups are extensively utilized as supports for many coastal structures, such as bridges, jetties, and oil production platforms. The problem of understanding the interaction effects within pile groups and predicting the breaking wave forces on them is considered in this paper, using experimental tests and machine learning-based predictive modeling. The restriction of previous studies on this important engineering problem is that the pile group arrangements considered are limited. Prediction methods are therefore developed only for specific pile group arrangements and do not incorporate the effect of the incident wave direction. In this study, to partially overcome this limitation, an extensive experimental investigation is conducted on 70 different pile group arrangements under six breaking wave conditions. Three pile group coefficients, characterized by the total, quasi-static, and dynamic forces, are introduced for a thorough assessment of the interaction effects within the pile group. First, the pile group coefficients for three basic arrangements (tandem, side-by-side, and staggered) are evaluated. The results reveal a sheltering effect in the tandem arrangement and an amplification effect in the side-by-side arrangement. However, the forces on the measured pile in the staggered arrangement resemble those on the isolated pile, with neither significant sheltering nor amplification effects observed. Then, the results for all arrangements highlight the significant effect of wave direction on the pile group coefficients for small inter-pile spacing. Finally, different machine learning algorithms are adopted to develop predictive models for the group coefficients. The XGBoost model demonstrates superior accuracy for predicting the total and quasi-static force coefficients, while the dynamic force coefficient remains challenging to predict accurately due to its stochastic nature.

Feature importance explanation of prosthetic valve endocarditis through machine learning via SHAP

Abstract Introduction Diagnosing prosthetic valve endocarditis poses a significant challenge in daily clinical practice. Accurately identifying these patients is vital due to its treatment implications. Clinical and imaging assessment of suspected cases yields a substantial number of interrelated variables, which can be analyzed through machine learning techniques in order to improve diagnostic accuracy. Feature ranking facilitates identification and interpretability of relevant clinical and imaging variables in this task. Purpose To utilize machine learning feature ranking and modeling in order to improve interpretable identification of patients with prosthetic valve endocarditis. Methods 89 cases defined by the endocarditis team and 173 controls were analyzed. We implemented the SHapley Additive exPlanations (SHAP) method to evaluate feature importance. 5-fold cross-validated Xgboost models were trained and independently tested with additive clinical, echocardiographic and 18F-FDG PET/CT data. Performance was evaluated through confusion matrices, F1-scores and AUCs. These were compared against conventional logistic regression of the Duke Criteria classification. Results Mean age was 59.6 ± 15.7, with 29.7% women. 14 clinical (demographic, laboratory and cultures), 2 echocardiographic, and 5 PET/CT features were included. The XGBoost model trained with all additive features demonstrated an AUC of 0.91±0.1 and outperformed simpler models as well as the Duke classification model (AUC=0.85). The clinical variables alone achieved an AUC of 0.85±0.11, while integration with echocardiography enhanced its performance to 0.90±0.04 (Fig. 1). Feature ranking demonstrated that antibiotic treatment duration, positive blood cultures, valve type, weight and CRP were the most relevant clinical variables, while the presence of vegetation or abscess, and abnormal uptake and uptake focality on PET/CT represented the most relevant imaging variables (Fig. 2). Conclusion Machine learning-based feature ranking and modeling may significantly enhance identification of patients with prosthetic valve endocarditis beyond standard clinical criteria. Feature selection offers interpretable performance in identifying these cases, model testing in clinical decision support may be warranted.Fig. 1:Comparative AUC Analysis.Fig. 2:SHAP feature importance.

Online Three-Dimensional Fuzzy Reinforcement Learning Modeling for Nonlinear Distributed Parameter Systems

Distributed parameter systems (DPSs) frequently appear in industrial manufacturing processes, with complex characteristics such as time–space coupling, nonlinearity, infinite dimension, uncertainty and so on, which is full of challenges to the modeling of the system. At present, most DPS modeling methods are offline. When the internal parameters or external environment of DPS change, the offline model is incapable of accurately representing the dynamic attributes of the real system. Establishing an online model for DPS that accurately reflects the real-time dynamics of the system is very important. In this paper, the idea of reinforcement learning is creatively integrated into the three-dimensional (3D) fuzzy model and a reinforcement learning-based 3D fuzzy modeling method is proposed. The agent improves the strategy by continuously interacting with the environment, so that the 3D fuzzy model can adaptively establish the online model from scratch. Specifically, this paper combines the deterministic strategy gradient reinforcement learning algorithm based on an actor critic framework with a 3D fuzzy system. The actor function and critic function are represented by two 3D fuzzy systems and the critic function and actor function are updated alternately. The critic function uses a TD (0) target and is updated via the semi-gradient method; the actor function is updated by using the chain derivation rule on the behavior value function and the actor function is the established DPS online model. Since DPS modeling is a continuous problem, this paper proposes a TD (0) target based on average reward, which can effectively realize online modeling. The suggested methodology is implemented on a three-zone rapid thermal chemical vapor deposition reactor system and the simulation results demonstrate the efficacy of the methodology.

Machine Learning-Based Profit Modeling for Credit Card Underwriting: Implications for Credit Risk

Accurate profit modeling is crucial for effective credit risk management in credit card underwriting. This study explores how machine learning methods can improve profit models in credit card underwriting, using data from a major Chinese bank as a case study. We employed three primary machine learning algorithms—Random Forest, Gradient Boosting Trees, and Neural Networks—to analyze a credit card customer dataset comprising 500,000 customers and 8 million transaction records. Model performance was evaluated using 10-fold cross-validation. The results show that the Gradient Boosting Trees model outperformed the others across all evaluation metrics. Specifically, the Gradient Boosting Trees model achieved an accuracy of 89.2%, a recall of 85.3%, an F1 score of 87.1%, and an AUC of 0.927, significantly surpassing the other models. Furthermore, the confusion matrix results indicated that the Gradient Boosting Trees model had the highest true positive and true negative values, and the lowest false positive and false negative values, further validating its superior classification performance. In summary, the Gradient Boosting Trees model demonstrated higher accuracy and stability in predicting profit in credit card underwriting, providing a robust basis for financial institutions to select appropriate machine learning models for credit risk management.

Changing the paradigm in heart failure: shifting from treatment to prevention.

Heart failure (HF) poses a major global health challenge with rising prevalence, significant morbidity and mortality, and substantial associated healthcare costs. With aging of the population and an increasing burden of comorbidities, the complex interplay between cardiovascular, kidney, and metabolic risk factors have been thrust into the spotlight and have broadened the traditional focus from HF treatment to an increased emphasis on prevention. In recognition of the evolving HF landscape, the American Heart Association released the PREVENT models which are comprehensive risk assessment tools that estimate 10- and 30-year risk of incident cardiovascular disease and its subtypes, including atherosclerotic cardiovascular disease (ASCVD) and, for the first time, HF. While it is an accurate risk estimation tool and represents a step forward in improving risk stratification for primary prevention of HF, there remain several limitations and unknowns like model performance across disaggregated racial and ethnic groups, the role of traditional ASCVD vs. HF-specific risk factors, HF prediction among those with known ASCVD, and the use of traditional regression techniques in lieu of potentially more powerful machine learning-based modeling approaches. Furthermore, it remains unclear how to optimize risk estimation in clinical care. The emergence of multiple novel pharmacological therapies that prevent incident HF, including sodium-glucose co-transporter 2 (SGLT2) inhibitors, glucagon-like peptide 1 (GLP1) receptor agonists, and nonsteroidal mineralocorticoid receptor antagonists (MRAs), highlights the importance of accurate HF risk prediction. To provide HF prevention with these effective but costly therapies, we must understand the optimal strategy in sequencing and combining these therapies and prioritize patients at highest risk. Such implementation requires both accurate risk stratification and a better understanding of how to communicate risk to patients and providers. This state-of-the-art review aims to provide a comprehensive overview of recent trends in HF prevention, including risk assessment, care management strategies, and emerging and novel treatments.

ResNet Modeling for 12 nm FinFET Devices to Enhance DTCO Efficiency

In this paper, a deep learning-based device modeling framework for design-technology co-optimization (DTCO) is proposed. A ResNet surrogate model is utilized as an alternative to traditional compact models, demonstrating high accuracy in both single-task (I–V or C–V) and multi-task (I–V and C–V) device modeling. Moreover, transfer learning is applied to the ResNet model, using the BSIM-CMG compact model for a 12 nm FinFET SPICE model as the pre-trained source. Through this approach, superior modeling accuracy and faster training speed are achieved compared to a ResNet surrogate model initialized with random weights, thereby meeting the rapid and efficient demands of the DTCO process. The effectiveness of the ResNet surrogate model in circuit simulation for 12 nm FinFET devices is demonstrated.