Machine learning with applications最新文献

{"title":"Modelling the effect of vaccination on the spread of COVID-19 via a novel evolutionary ensemble learning algorithm","authors":"Mohammad Hassan Tayarani Najaran","doi":"10.1016/j.mlwa.2025.100720","DOIUrl":"10.1016/j.mlwa.2025.100720","url":null,"abstract":"<div><div>The spread of the COVID-19 disease has caused a lot of problems for every country around the world. To curb the pandemic, governments have issued various policies, including vaccination. Depending on the percentage of the vaccinated population, the pandemic responds differently to the policies. This paper proposes a modelling algorithm that takes as input the percentage of the vaccinated population and the policies taken by governments and generates as output a prediction of the number of newly infected cases. Then, this model is used as the fitness function in an optimisation algorithm, which for a population with a certain percentage of vaccinated people, searches through the set of policies and finds the best set of policies that minimises the cost to society and the number of infected people. To build the model, an ensemble learning algorithm is proposed, which is a combination of different learning algorithms. In this algorithm, an evolutionary diversifier algorithm is proposed to generate the base learners. The algorithm chooses different subsets of features for each base learner to maximise diversity among them. Then, an evolutionary process is adopted to choose from the base learners a subset that optimises the prediction accuracy of the model. The proposed algorithms are tested on a well-known data set about government policies, the percentage of the population vaccinated, and the number of infected cases. Experimental studies suggest better performance for the proposed ensemble learning algorithm compared to existing ones. Multi-objective optimisation of the policies is also proposed and tested on the model, and the results are presented in this paper.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"21 ","pages":"Article 100720"},"PeriodicalIF":4.9,"publicationDate":"2025-08-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144829244","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Bayesian inversion supervised learning framework for the enzyme activity in graphene field-effect transistors","authors":"Ehsan Khodadadian ,&nbsp;Samaneh Mirsian ,&nbsp;Shahrzad Shashaani ,&nbsp;Maryam Parvizi ,&nbsp;Amirreza Khodadadian ,&nbsp;Peter Knees ,&nbsp;Wolfgang Hilber ,&nbsp;Clemens Heitzinger","doi":"10.1016/j.mlwa.2025.100718","DOIUrl":"10.1016/j.mlwa.2025.100718","url":null,"abstract":"<div><div>Graphene Field-Effect Transistors (GFETs) are gaining prominence in enzyme detection due to their exceptional sensitivity, rapid response, and capability for real-time monitoring of enzymatic reactions. Among different catalytic systems, heme-based peroxidase enzymes such as horseradish peroxidase (HRP), and heme molecules, which can exhibit peroxidase-like activity, are noteworthy due to their significant catalytic behavior. GFETs effectively monitor and detect these enzymatic reactions by observing alterations in their electrical properties. In this study, we present a computational framework designed to determine key enzymatic parameters, including the enzyme turnover number and the Michaelis–Menten constant. Utilizing experimental reaction rate data obtained from the GFET electrical response, we apply Bayesian inversion models to estimate these parameters accurately. Additionally, we develop a novel deep neural network (multilayer perceptron) to predict enzyme behavior under various chemical and environmental conditions. The performance of this coupled computational model is compared against standard machine learning and Bayesian inversion techniques to validate its efficiency and accuracy. We present a pseudocode to explain the implementation of machine learning Bayesian inversion framework.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"21 ","pages":"Article 100718"},"PeriodicalIF":4.9,"publicationDate":"2025-08-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144829243","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhanced early detection of dysarthric speech disabilities using stacking ensemble deep learning model","authors":"Jagat Chaitanya Prabhala ,&nbsp;Ravi Ragoju ,&nbsp;Venkatanareshbabu Kuppili ,&nbsp;Christophe Chesneau","doi":"10.1016/j.mlwa.2025.100721","DOIUrl":"10.1016/j.mlwa.2025.100721","url":null,"abstract":"<div><div>Communication disorders, particularly dysarthria, significantly impact individuals by impairing their speech clarity, social interactions, and overall well-being. Early and accurate detection is crucial to enable timely intervention and improve speech therapy outcomes. This study introduces Adaptive Dysarthric Speech Disability Detection using Stacked Ensemble Deep Learning (ADSDD-SEDL), an innovative ensemble-based deep-learning framework for dysarthria detection. The proposed model integrates three deep learning architectures—Multi-Head Attention-based Long Short-Term Memory (MHALSTM), Deep Belief Network (DBN), and Time-Delay Neural Network (TDNN)—within a stacked ensemble model. Unlike conventional stacking methods that use fixed meta-classifiers, this study employs a Genetic Algorithm (GA)-based optimization strategy to dynamically determine optimal weight contributions of the base models, enhancing classification robustness and adaptability.</div><div>The preprocessing pipeline converts speech signals from the time domain to the frequency domain by using a Short-Time Fourier Transform (STFT). Mel-Frequency Cepstral Coefficients (MFCCs) were extracted to capture the key spectral characteristics. Each base model underwent independent training, and the GA optimized the ensemble by evolving an adaptive weight distribution instead of relying on predefined fusion methods. Extensive simulations and hyperparameter tuning confirmed that the GA-optimized ADSDD-SEDL technique significantly improved detection efficiency over traditional ensemble approaches. These findings underscore the advantages of evolutionary optimization in refining speech disorder classification models. This scalable and adaptive model offers a valuable tool for healthcare professionals, enabling precise and automated early diagnosis of dysarthria. Future research could explore alternative evolutionary algorithms, reinforcement learning techniques, and hybrid deep learning approaches to enhance speech disorder classification.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"21 ","pages":"Article 100721"},"PeriodicalIF":4.9,"publicationDate":"2025-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144809565","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Configurable Intrinsic Curiosity Module for a Testbed for Developing Intelligent Swarm UAVs","authors":"Jawad Mahmood ,&nbsp;Muhammad Adil Raja ,&nbsp;John Loane ,&nbsp;Fergal McCaffery","doi":"10.1016/j.mlwa.2025.100714","DOIUrl":"10.1016/j.mlwa.2025.100714","url":null,"abstract":"<div><div>This paper introduces an Intrinsic Curiosity Module (ICM) based Reinforcement Learning (RL) framework for swarm Unmanned Aerial Vehicles (UAVs) target tracking, leveraging the actor–critic architecture to control the roll, pitch, yaw, and throttle motions of UAVs. A key challenge in RL-based UAV coordination is the delayed reward problem, which hinders effective learning in dynamic environments. Existing UAV testbeds rely primarily on extrinsic rewards and lack mechanisms for adaptive exploration and efficient UAV coordination. To address these limitations, we propose a testbed that integrates an ICM with the Asynchronous Advantage Actor-Critic (A3C) algorithm for tracking UAVs. It incorporates the Self-Reflective Curiosity-Weighted (SRCW) hyperparameter tuning mechanism for the ICM, which adaptively modifies hyperparameters based on the ongoing RL agent’s performance. In this testbed, the target UAV is guided by the Advantage Actor-Critic (A2C) model, while a swarm of two tracking UAVs is controlled by using the A3C-ICM approach. The proposed framework facilitates real-time autonomous coordination among UAVs within a simulated environment. This system is developed using the FlightGear flight simulator and the JSBSim Flight Dynamics Model (FDM), which enables dynamic simulations and continuous interaction between UAVs. Experimental results demonstrate that the tracking UAVs can effectively coordinate and maintain precise paths even under complex conditions.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"21 ","pages":"Article 100714"},"PeriodicalIF":4.9,"publicationDate":"2025-08-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144766641","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A fine-tuned deep learning model for detecting Japanese beetles in soybeans using unmanned aircraft systems (UAS) and mobile imaging","authors":"Ivan Grijalva ,&nbsp;H. Braden Adams ,&nbsp;Brian McCornack","doi":"10.1016/j.mlwa.2025.100711","DOIUrl":"10.1016/j.mlwa.2025.100711","url":null,"abstract":"<div><div>Since the early 1900s, the Japanese beetle (<em>Popillia japonica</em>, Newman) has invaded soybean crops in the U.S. It has emerged as a significant economic threat due to its habit of defoliating plants, often leaving them skeletal and reducing yields. The conventional approach for monitoring Japanese beetles uses visual assessments and sweep counts, which are impractical for larger soybean acreages. Furthermore, frequent manual sampling demands labor and time resources that could otherwise be allocated to enhancing soybean production. To address this challenge, we fine-tuned a deep learning model capable of automatically detecting Japanese beetles using images, thereby improving the monitoring process. The YOLOv8s model can detect Japanese beetles 87.90 % of the time on images collected from mobile devices and unmanned aircraft systems (UAS) at certain distances. The model was deployed in a web application as a prototype platform to understand the capabilities of deep learning in pest monitoring. This web application is a server that autonomously analyzes images captured by mobile devices and UAS to detect and count beetles in the soybean canopy. This study aimed to transform the traditional method of pest monitoring in soybean production by transitioning to a digital monitoring system.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"21 ","pages":"Article 100711"},"PeriodicalIF":0.0,"publicationDate":"2025-07-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144694594","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Modeling of settlement of shallow-founded rocking structures using explainable physics-guided machine learning","authors":"Sivapalan Gajan,&nbsp;Christopher Kantor","doi":"10.1016/j.mlwa.2025.100702","DOIUrl":"10.1016/j.mlwa.2025.100702","url":null,"abstract":"<div><div>Rocking foundation is an unorthodox seismic design philosophy of structures that enhances the performance of structures by absorbing and dissipating seismic energy into soil. This paper examines the application of physics-guided machine learning (PGML) technique to model the settlement of shallow-founded rocking structures during earthquake loading. An approximate physics-based model (PBM) is derived for rocking-induced total settlement as a function of critical contact area ratio and cumulative rotation of the foundation. The output of the PBM is fed as an additional input feature to machine learning (ML) algorithms to develop PGML models. The performances of PGML models are compared with the performances of purely data-driven ML models, the PBM outputs, and results obtained from an empirical relationship. To shed light on the explainability of ML and PGML models, Shapley Additive Explanations (SHAP values) are used to decipher and interpret the model predictions and their dependency on input features. It is found that PGML models, especially physics-guided gradient boosting and random forest regression, improve the prediction accuracy when compared to their purely data-driven ML counterparts by combining the knowledge extracted from experimental data with the mechanics of the problem considered. SHAP analysis reveals that the PGML model predictions and their dependency on input features are consistent with the existing domain knowledge, and that the inclusion of physics in PGML models help improve the prediction accuracy, especially in cases where other input features fail to capture the combined complex interaction among the variables involved.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"21 ","pages":"Article 100702"},"PeriodicalIF":0.0,"publicationDate":"2025-07-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144678937","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-objective deep learning: Taxonomy and survey of the state of the art","authors":"Sebastian Peitz,&nbsp;Sèdjro Salomon Hotegni","doi":"10.1016/j.mlwa.2025.100700","DOIUrl":"10.1016/j.mlwa.2025.100700","url":null,"abstract":"<div><div>Simultaneously considering multiple objectives in machine learning has been a popular approach for several decades, with various benefits for multi-task learning, the consideration of secondary goals such as sparsity, or multicriteria hyperparameter tuning. However – as multi-objective optimization is significantly more costly than single-objective optimization – the recent focus on deep learning architectures poses considerable additional challenges due to the very large number of parameters, strong nonlinearities and stochasticity. On the other hand considering multiple criteria in deep learning presents many benefits, such as the just-mentioned multi-task learning, the consideration of performance versus adversarial robustness, or a more interpretable way for interactively adapting to changing preferences. This survey covers recent advancements in the area of multi-objective deep learning. We introduce a taxonomy of existing methods – based on the type of training algorithm as well as the decision maker’s needs – before listing recent advancements, and also successful applications. All three main learning paradigms supervised learning, unsupervised learning and reinforcement learning are covered, and we also address the recently very popular area of generative modeling. With a focus on the advantages and disadvantages of the existing training algorithms, this survey is formulated from an optimization perspective rather than organizing according to different learning paradigms or application areas.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"21 ","pages":"Article 100700"},"PeriodicalIF":0.0,"publicationDate":"2025-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144662492","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Improving neural network training using dynamic learning rate schedule for PINNs and image classification","authors":"Veerababu Dharanalakota ,&nbsp;Ashwin Arvind Raikar ,&nbsp;Prasanta Kumar Ghosh","doi":"10.1016/j.mlwa.2025.100697","DOIUrl":"10.1016/j.mlwa.2025.100697","url":null,"abstract":"<div><div>Training neural networks can be challenging, especially as the complexity of the problem increases. Despite using wider or deeper networks, training them can be a tedious process, especially if a wrong choice of the hyperparameter is made. The learning rate is one of such crucial hyperparameters, which is usually kept static during the training process. Learning dynamics in complex systems often requires a more adaptive approach to the learning rate. This adaptability becomes crucial to effectively navigate varying gradients and optimize the learning process during the training process. In this paper, a dynamic learning rate scheduler (DLRS) algorithm is presented that adapts the learning rate based on the loss values calculated during the training process. Experiments are conducted on problems related to physics-informed neural networks (PINNs) and image classification using multilayer perceptrons and convolutional neural networks, respectively. The results demonstrate that the proposed DLRS accelerates training and improves stability.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"21 ","pages":"Article 100697"},"PeriodicalIF":0.0,"publicationDate":"2025-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144711963","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A comprehensive overview of remaining useful life prediction: From traditional literature review to scientometric analysis","authors":"Yitong Liu,&nbsp;Jiarui Wen,&nbsp;Guoqiang Wang","doi":"10.1016/j.mlwa.2025.100704","DOIUrl":"10.1016/j.mlwa.2025.100704","url":null,"abstract":"<div><div>The increasing complexity of industrial systems has heightened the need for precise Remaining Useful Life (RUL) predictions. This paper provides a comprehensive overview of RUL prediction methods, processes, datasets, and tools, alongside insights from scientometric analysis. We categorize RUL prediction methods into model-based, data-driven, and hybrid methods, detailing key models and their applications. Model-based methods utilize physical failure mechanisms, enhancing interpretability, but with limited adaptability for complex systems. Data-driven methods use machine learning to extensive datasets, boosting adaptability but often at the cost of explainability. Hybrid methods aim to strike a balance between these strengths, offering both accuracy and flexibility across various applications. In addition, we outline the RUL prediction process, from data acquisition to implementation, and introduce the primary datasets used in studies of engines, bearings, and batteries as well as commonly used programming frameworks and open-source libraries. We also performed a scientometric analysis of 3442 articles from the Web of Science database, examining research areas, prominent research groups, and emerging trends. Our findings offer a comprehensive overview of the RUL prediction field, outlining its current trends and highlighting potential directions for future research.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"21 ","pages":"Article 100704"},"PeriodicalIF":0.0,"publicationDate":"2025-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144687081","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Predictive modelling of graphene-enhanced greases using classical feedback control and quantum kernel regression","authors":"Ethan Stefan-Henningsen ,&nbsp;Amirkianoosh Kiani","doi":"10.1016/j.mlwa.2025.100705","DOIUrl":"10.1016/j.mlwa.2025.100705","url":null,"abstract":"<div><div>This paper investigates two predictive modeling approaches for estimating the thermal and tribological performance of graphene-enhanced greases, aiming to reduce reliance on protracted endurance tests. Seven grease formulations with varying graphene concentrations (0–4 wt%) were prepared and tested under a uniform load to capture temperature evolution, wear scar area and coefficient of friction. A classical piecewise regression model, augmented by a Linear Quadratic Regulator (LQR), leverages feedback control to correct temperature predictions and subsequently estimate wear using a polynomial fit. This framework demonstrated high accuracy in tracking transient thermal behaviour, maintaining temperature deviations within ±1 °C of measured data. In parallel, a quantum-classical hybrid model employs a fidelity-based quantum kernel with support vector regression. By encoding partial early-cycle temperature measurements (e.g., from 30 to 120s) into a higher-dimensional Hilbert space, the quantum approach captures subtle nonlinearities and yields strong correlations for both final temperature and wear scar area. Moreover, consistent performance on IBM Quantum models with realistically simulated noise underscores the model’s potential for practical industrial implementation. Collectively, these results confirm the viability of advanced computational tools, both classical and quantum, for rapid, data-driven lubricant assessments. They highlight opportunities to optimize graphene content while minimizing costly trial and error testing.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"21 ","pages":"Article 100705"},"PeriodicalIF":0.0,"publicationDate":"2025-07-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144711964","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}