Create Machine Learning Models Research Articles

Preprocessing and Modeling Approach for Gearbox Pitting Severity Prediction under Unseen Operating Conditions and Fault Severities

Gear pitting is a common gearbox failure mode that can lead to unplanned machine downtime, inefficient power transmission and a higher risk of sudden catastrophic failure. Consequently, there is strong incentive to create machine learning models that are capable of detecting and quantifying the severity of gearbox pitting faults. The performance of machine learning models is however highly dependent on the availability of training data and since training data for a wide variety of different operating conditions and fault severities is rarely available in practice, machine learning models must be designed to be robust to unseen operating conditions and fault severities. Furthermore, models should be capable of identifying data outside of the training data distribution and adjusting the confidence in a prediction accordingly. This work presents a strategy for pitting severity estimation in gearboxes under unseen operating conditions and fault severities in response to the PHM North America 2023 Conference Data Challenge. The strategy includes the design of dedicated validation sets for quantifying model performance on unseen data, an investigation into the most appropriate preprocessing methods, and a specialized convolutional neural network with an integrated out of distribution detection model for identifying samples from foreign operating conditions and fault severities. The results show that the best models are capable of some generalization to unseen operating conditions, but the generalization to unseen pitting severities is more challenging.

Creating machine learning models that interpretably link systemic inflammatory index, sex steroid hormones, and dietary antioxidants to identify gout using the SHAP (SHapley Additive exPlanations) method.

The relationship between systemic inflammatory index (SII), sex steroid hormones, dietary antioxidants (DA), and gout has not been determined. We aim to develop a reliable and interpretable machine learning (ML) model that links SII, sex steroid hormones, and DA to gout identification. The dataset we used to study the relationship between SII, sex steroid hormones, DA, and gout was from the National Health and Nutrition Examination Survey (NHANES). Six ML models were developed to identify gout by SII, sex steroid hormones, and DA. The seven performance discriminative features of each model were summarized, and the eXtreme Gradient Boosting (XGBoost) model with the best overall performance was selected to identify gout. We used the SHapley Additive exPlanation (SHAP) method to explain the XGBoost model and its decision-making process. An initial survey of 20,146 participants resulted in 8,550 being included in the study. Selecting the best performing XGBoost model associated with SII, sex steroid hormones, and DA to identify gout (male: AUC: 0.795, 95% CI: 0.746- 0.843, accuracy: 98.7%; female: AUC: 0.822, 95% CI: 0.754- 0.883, accuracy: 99.2%). In the male group, The SHAP values showed that the lower feature values of lutein + zeaxanthin (LZ), vitamin C (VitC), lycopene, zinc, total testosterone (TT), vitamin E (VitE), and vitamin A (VitA), the greater the positive effect on the model output. In the female group, SHAP values showed that lower feature values of E2, zinc, lycopene, LZ, TT, and selenium had a greater positive effect on model output. The interpretable XGBoost model demonstrated accuracy, efficiency, and robustness in identifying associations between SII, sex steroid hormones, DA, and gout in participants. Decreased TT in males and decreased E2 in females may be associated with gout, and increased DA intake and decreased SII may reduce the potential risk of gout.

Financial Statement Text Information Mining and Key Information Extraction Model Construction

Financial statement text information mining and key information extraction model design are critical areas of research that aim to use advanced computational approaches to extract important insights from textual data contained in financial documents. In this work, they look at methodologies, techniques, and applications that combine natural language processing (NLP) and machine learning to automate financial statement interpretation. To lay the groundwork for the research, researchers first conduct a thorough examination of existing literature in interdisciplinary domains such as computational linguistics, information retrieval, and finance. Building on insights from earlier studies, they design and use unique NLP approaches, such as named entity identification, syntactic parsing, sentiment analysis, and topic modelling, to extract essential financial metrics from textual data. Additionally, they create machine learning models that are suited to the peculiarities of financial terminology and reporting standards, combining domain-specific knowledge with linguistic experience to improve accuracy and reliability. They demonstrate the efficacy and scalability of the technique in automating the extraction of crucial financial information, such as revenue trends, cost patterns, and risk factors, through rigorous testing on real-world financial data. These results highlight the transformative power of natural language processing and machine learning in financial analysis, providing stakeholders in finance and accounting with actionable intelligence for informed decision-making, risk assessment, and compliance monitoring. By bridging the gap between computational linguistics and financial analysis, this study advances financial text analysis and provides the framework for future research and innovation in this emerging field.

Early Classification of Gram-negative Bacteria with Colony Imaging and Deep Learning without Coding Experience

Bacterial colony morphology is the first step in classifying bacterial species during the microbial identification process. It is very important to assess the morphology of bacterial colonies in a preliminary screening process to largely reduce the scope of possible bacteria species and increase work productivity in clinical bacteriology by making later identification more specific. However, making a decision about this topic requires sufficient clinical laboratory expertise. Teachable Machine® is a rapid, easy-to-use, web-based tool accessible to everyone that is used to create machine learning models. In this study, the performance of Teachable Machine® was assessed for cheap, rapid and practical identification of enteric and non-fermenting bacteria frequently isolated in microbiology laboratories. A total of 1202 colony images were used to train and validate the network's diagnostic performance. Additionally, 80 representative test images were used to assess performance. Level 1 was defined as E. coli-K. pneumonia, Level 2 was defined as P. aeruginosa-A. baumannii, Level 3 was defined as enteric bacteria-non-fermenting bacteria and Level 4 was defined as differentiating these four pathogens from each other. Mean accuracy of Teachable Machine® for the defined classes was 96.7%, 94.1%, 94.3%, and 90.3% for Levels 1, 2, 3, and 4, respectively. General accuracy for classification of the 80 representative colonies was 82.5% and the hit rates were 85.0%, 100%, 75.0%, and 70.0% for E. coli, K. pneumoniae, P. aeruginosa and A. baumannii, respectively. This cost-effective bacterial identification system, supported by deep learning, will be an important pioneer for a variety of applications in clinical microbiology by reducing the identification process by a significant degree and automating identification of colonies without requiring a specialist.

Machine learning models for COVID-19 diagnosis based on medical images and audio

The COVID-19 pandemic presents a significant danger to human health, with far-reaching consequences for the global economy and political dynamics. It presents complex challenges in terms of rapid and accurate diagnosis, where machine learning holds potential to enhance diagnostic speed and precision while reducing time and resource burdens. Consequently, this research employs CT images and cough audio recordings as training data to create machine learning models for data classification, with the goal of aiding in the diagnosis of COVID-19. Using the Edge Impulse platform, a Convolutional Neural Network, MobileNetV2, is customized for efficient image recognition. On the audio front, the preprocessing phase encompassed three distinct feature extraction techniques: Mel Frequency Cepstral Coefficients, Mel-Filterbank Energy (MFE), and Mel spectrogram. Subsequently, model frameworks were meticulously adjusted to accommodate the classification requirements. The results of this effort are highly encouraging. In the domain of CT image recognition, the top-performing model achieved a remarkable accuracy of 93.98%. In the concurrent task of categorizing cough audio data, the best performance reached 81.8%. These findings underscore the capability of this approach as an effective supplementary tool for medical diagnostics. In the face of COVID-19s persisting impact, such machine learning advancements could significantly aid in swift and reliable diagnoses, ultimately contributing to the global battle against the pandemic.

Machine learning assisted exploration of the influential parameters on the PLGA nanoparticles

Poly (lactic-co-glycolic acid) (PLGA)-based nanoparticles (NPs) are widely investigated as drug delivery systems. However, despite the numerous reviews and research papers discussing various physicochemical and technical properties that affect NP size and drug loading characteristics, predicting the influential features remains difficult. In the present study, we employed four different machine learning (ML) techniques to create ML models using effective parameters related to NP size, encapsulation efficiency (E.E.%), and drug loading (D.L.%). These parameters were extracted from the different literature. Least Absolute Shrinkage and Selection Operator was used to investigate the input parameters and identify the most influential features (descriptors). Initially, ML models were trained and validated using tenfold validation methods, and subsequently, next their performances were evaluated and compared in terms of absolute error, mean absolute, error and R-square. After comparing the performance of different ML models, we decided to use support vector regression for predicting the size and E.E.% and random forest for predicting the D.L.% of PLGA-based NPs. Furthermore, we investigated the interactions between these target variables using ML methods and found that size and E.E.% are interrelated, while D.L.% shows no significant relationship with the other targets. Among these variables, E.E.% was identified as the most influential parameter affecting the NPs' size. Additionally, we found that certain physicochemical properties of PLGA, including molecular weight (Mw) and the lactide-to-glycolide (LA/GA) ratio, are the most determining features for E.E.% and D.L.% of the final NPs, respectively.

Predicting the crack repair rate of self-healing concrete using soft-computing tools

To tackle the challenges of cracking and insufficient durability in traditional concrete, researchers are exploring self-healing concrete (SHC) as a potential alternative material. Nevertheless, doing laboratory studies can incur significant expenses and consume considerable time. Therefore, utilizing machine learning (ML) algorithms can contribute to advancing improved predictions for self-healing concrete. This study aimed to create ML models, specifically adaptive boosting (AB), artificial neural network (ANN), and gradient boosting (GB), to forecast and evaluate the repair rate of the cracked region in SHC that incorporates fibers and bacteria in their compositions. To further evaluate the importance of inputs, RReliefF analysis was performed. The findings indicated that the AB algorithm outperformed GB and ANN algorithms by achieving a coefficient of determination (R2) value of 0.987, a mean absolute error (MAE) value of 0.001 mm, and a root mean square error (RMSE) value of 0.026 mm. The GB approach yielded an R2 value of 0.962, an MAE value of 0.035 mm, and an RMSE value of 0.044 mm. Similarly, the ANN approach yielded an R2 value of 0.943, an MAE value of 0.040 mm, and an RMSE value of 0.054 mm. These results demonstrate that all AB, GB, and ANN algorithms outperformed in terms of prediction accuracy and model fit. Hence, the application of these models can be employed to construct and verify advanced SHC compositions that rely on polymeric fibers and bacteria, aiming to achieve superior performance. However, comparing the outcomes of developed models revealed that the AB model accuracy is higher than the GB and ANN models for predicting the area repair rate of cracks.

ABDpred: Prediction of active antimicrobial compounds using supervised machine learning techniques.

Discovery of new antibiotics is the need of the hour to treat infectious diseases. An ever-increasing repertoire of multidrug-resistant pathogens poses an imminent threat to human lives across the globe. However, the low success rate of the existing approaches and technologies for antibiotic discovery remains a major bottleneck. In silico methods like machine learning (ML) deem more promising to meet the above challenges compared with the conventional experimental approaches. The goal of this study was to create ML models that may be used to successfully predict new antimicrobial compounds. In this article, we employed eight different ML algorithms namely, extreme gradient boosting, random forest, gradient boosting classifier, deep neural network, support vector machine, multilayer perceptron, decision tree, and logistic regression. These models were trained using a dataset comprising 312 antibiotic drugs and a negative set of 936 non-antibiotic drugs in a five-fold cross validation approach. The top four ML classifiers (extreme gradient boosting, random forest, gradient boosting classifier and deep neural network) were able to achieve an accuracy of 80 per cent and above during the evaluation of testing and blind datasets. We aggregated the top performing four models through a soft-voting technique to develop an ensemble-based ML method and incorporated it into a freely accessible online prediction server named ABDpred ( http://clinicalmedicinessd.com.in/abdpred/ ).

Forecasting hourly electricity generation by a solar power plant using machine learning algorithms

Relevance. The need to develop energy-saving approaches through the use of data mining tools to improve the efficiency of management decision-making and more optimal use of energy resources. Forecasting the amount of electric energy generated by a solar power plant will allow optimal electricity distribution in decentralized systems. Information about the amount of electricity generated by a solar power plant and transmitted to a grid at every hour will allow planning the use of generated electricity and its distribution more reasonably. Also, the presence of a reliable forecast will allow embedding a predictive model into the micro-grid management subsystem. This will facilitate the integration of centralized electrical networks and distributed generation facilities. Aim. To analyze scientific papers containing proposals to improve the accuracy of determining the amount of electricity generated by solar power plants; to create machine learning models that allow you to create a short-term forecast of electricity generation by a solar power plant. Objects. Solar power plant named after A.A. Vlaznev (Sakmarskaya SPP), Orenburg region, Orsk, Oktyabrsky district, geographical coordinates: 51.266023, 58.474689. Methods. Analytical method, methods of mathematical statistics, methods of machine learning, complex generalization of scientific achievements and practical experience in the use of data processing tools in the tasks of forecasting electricity generation by solar power plants. Results. The authors have carried out the review of literature sources reflecting the results of using modern data mining tools in predicting the magnitude of electricity generation by solar power plants. The paper considers various approaches to forecasting electricity generation at solar power plants. The analysis of factors used in forecasting is carried out. The authors obtained the results of theoretical and applied nature. The results consist in recommendations on using exogenous variables in predicting power generation at SPP, as well as some machine learning algorithms in construction of predictive models. These recommendations were obtained in generalizing the results of the applied research.

A Survey on Explainable Artificial Intelligence for Cybersecurity

The black-box nature of artificial intelligence (AI) models has been the source of many concerns in their use for critical applications. Explainable Artificial Intelligence (XAI) is a rapidly growing research field that aims to create machine learning models that can provide clear and interpretable explanations for their decisions and actions. In the field of network cybersecurity, XAI has the potential to revolutionize the way we approach network security by enabling us to better understand the behavior of cyber threats and to design more effective defenses. In this survey, we review the state of the art in XAI for cybersecurity in network systems and explore the various approaches that have been proposed to address this important problem. The review follows a systematic classification of network-driven cybersecurity threats and issues. We discuss the challenges and limitations of current XAI methods in the context of cybersecurity and outline promising directions for future research.

Use of Natural Language Processing to Identify Sexual and Reproductive Health Information in Clinical Text.

This study aimed to enable clinical researchers without expertise in natural language processing (NLP) to extract and analyze information about sexual and reproductive health (SRH), or other sensitive health topics, from large sets of clinical notes. (1) We retrieved text from the electronic health record as individual notes. (2) We segmented notes into sentences using one of scispaCy's NLP toolkits. (3) We exported sentences to the labeling application Watchful and annotated subsets of these as relevant or irrelevant to various SRH categories by applying a combination of regular expressions and manual annotation. (4) The labeled sentences served as training data to create machine learning models for classifying text; specifically, we used spaCy's default text classification ensemble, comprising a bag-of-words model and a neural network with attention. (5) We applied each model to unlabeled sentences to identify additional references to SRH with novel relevant vocabulary. We used this information and repeated steps 3 to 5 iteratively until the models identified no new relevant sentences for each topic. Finally, we aggregated the labeled data for analysis. This methodology was applied to 3,663 Child Neurology notes for 971 female patients. Our search focused on six SRH categories. We validated the approach using two subject matter experts, who independently labeled a sample of 400 sentences. Cohen's kappa values were calculated for each category between the reviewers (menstruation: 1, sexual activity: 0.9499, contraception: 0.9887, folic acid: 1, teratogens: 0.8864, pregnancy: 0.9499). After removing the sentences on which reviewers did not agree, we compared the reviewers' labels to those produced via our methodology, again using Cohen's kappa (menstruation: 1, sexual activity: 1, contraception: 0.9885, folic acid: 1, teratogens: 0.9841, pregnancy: 0.9871). Our methodology is reproducible, enables analysis of large amounts of text, and has produced results that are highly comparable to subject matter expert manual review.

Predictions of Programmed Cell Death Ligand 1 Blockade Therapy Success in Patients with Non-Small-Cell Lung Cancer

Lung cancer is responsible for the most cancer deaths worldwide, with non-small-cell lung cancer (NSCLC) making up 80% of cases. Some genetic factors leading to NSCLC development include genetic mutations and Programmed Cell Death Ligand 1 (PD-L1) expression. PD-L1 proteins are targeted in an NSCLC treatment called PD-L1 blockade therapy (immune therapy). However, this treatment is effective in a low percentage of patients. This study aimed to create machine learning models to use features, like the number of mutations and the number of PD-L1 proteins in cancer cells, along with others, to predict whether a patient will receive clinical benefits from immune therapy. This was carried out by downloading and merging datasets from cbioportal.org to create a sample size for the model. Features that were highly correlated with clinical benefits were identified. Three machine learning models (Gaussian naïve Bayes, decision tree, and logistic regression) were created using these features to predict clinical benefits in patients, and each model’s accuracy was evaluated. All three models had accuracy rates between 55 and 85%, with two of the models averaging an accuracy rate of around 75%. Doctors can use these models to more accurately predict whether immune therapy treatment is likely to work in a patient before prescribing it to them.

Easy-to-Use and Customizable Data Science Tool for Predictive Maintenance in Manufacturing

Even if many manufacturing companies have attempted to utilize data analytics for predictive maintenance but have not been able to do so. This is because predictive with data requires both knowledge of equipment in production site and knowledge of data science, and such workers are scarce. Another problem is that it takes time and effort to apply the created machine learning models on site and to develop applications for monitoring the diagnosis results. Therefore, we developed “MELSOFT MaiLab”. The application has an intuitive interface and functions to automatically create machine learning models, making it possible for workers with no knowledge of data science to easily create machine learning models for predictive maintenance. The application also has functions for deploying the created machine learning models to the production site and monitoring the diagnostic results.

Using Machine Learning Prediction to Create a 15-question IPV Measurement Tool.

Domestic violence, especially intimate partner violence (IPV), is an important issue worldwide, especially in India. Those that experience it may not always be able to come forward or have access to the required social support to act against it. We use National Family Health Survey data (n = 66,013 women) to create machine learning models which can predict IPV instances with a recall of 78%. We use the top 15 best predicting questions that avoid sensitive issues to create a field tool that frontline health workers can use to identify women with a high risk of IPV and provide the support they need.

Computer Vision Identifies Recurrent and Nonrecurrent Ductal Carcinoma in Situ Lesions with Special Emphasis on African-American Women

Although nonrecurrent and recurrent forms of ductal carcinoma in situ (DCIS) of the breast are observed, no evidence-based test can make this distinction. The current retrospective case-control study used archival DCIS samples stained with anti-phospho-Ser226-glucose transporter type 1 and anti-phosphofructokinase type L antibodies. Immunofluorescence micrographs were used to create machine learning models of recurrent and nonrecurrent biomarker patterns, which were evaluated in cross-validation studies. Clinical performance was assessed by holdout studies using patients whose data were not used in training. Micrographs were stratified according to the recurrence probability of each image. Recurrent patients were defined by at least one image with a probability of recurrence ≥98%, whereas nonrecurrent patients had none. These studies found no false-negatives, identified true-positives, and uniquely identified true-negatives. Roughly 20% of the microscope fields of recurrent lesions were computationally recurrent. Strong prognostic results were obtained for both white and African-American women. This machine tool provides the first means to accurately predict recurrent and nonrecurrent patient outcomes. Data indicate that at least some false-positive findings were true-positive findings that benefited from surgical intervention. The intracellular locations of phospho-Ser226-glucose transporter type 1 and phosphofructokinase type L likely participate in cancer recurrences by accelerating glucose flux, a key feature of the Warburg effect.

Development of a data-driven digital phenotype profile of distress experience of healthcare workers during COVID-19 pandemic

Background and objective:Due to the constraints of the COVID-19 pandemic, healthcare workers have reported acting in ways that are contrary to their moral values, and this may result in moral distress. This paper proposes the novel digital phenotype profile (DPP) tool, developed specifically to evaluate stress experiences within participants. The DPP tool was evaluated using the COVID-19 VR Healthcare Simulation of Stress Experience (HSSE) dataset (NCT05001542), which is composed of passive physiological signals and active mental health questionnaires. The DPP tool focuses on correlating electrocardiogram, respiration, photoplethysmography, and galvanic skin response with moral injury outcome scale (Brief MIOS).Methods:Data-driven techniques are encompassed to develop a tool for robust evaluation of distress among participants. To accomplish this, we applied pre-processing techniques which involved normalization, data sanitation, segmentation, and windowing. During feature analysis, we extracted domain-specific features, followed by feature selection techniques to rank the importance of the feature set. Prior to classification, we employed k-means clustering to group the Brief MIOS scores to low, moderate, and high moral distress as the Brief MIOS lacks established severity cut-off scores. Support vector machine and decision tree models were used to create machine learning models to predict moral distress severities.Results:Weighted support vector machine with leave-one-subject-out-cross-validation evaluated the separation of the Brief MIOS scores and achieved an average accuracy, precision, sensitivity, and F1 of 98.67%, 98.83%, 99.44%, and 99.13%, respectively. Various machine learning ablation tests were performed to support our results and further enhance the understanding of the predictive model.Conclusion:Our findings demonstrate the feasibility to develop a DPP tool to predict distress experiences using a combination of mental health questionnaires and passive signals. The DPP tool is the first of its kind developed from the analysis of the HSSE dataset. Additional validation is needed for the DPP tool through replication in larger sample sizes.

The Early Detection of Neurodegenerative diseases initiative: an international and multidisciplinary effort for transforming the early detection of dementia‐causing diseases

AbstractBackgroundAround 50 million people have dementia worldwide, with nearly 10 million new cases every year. Diagnosis is complex and often relies on expensive and invasive measures, with most patients accessing medical support when they already experience symptoms.MethodThe Early Detection of Neurodegenerative diseases (EDoN) initiative, spearheaded by Alzheimer’s Research UK, brings together over 60 experts from 49 universities, research projects, patient cohorts and technology providers to create machine learning models to detect the earliest stages of dementia‐causing diseases. EDoN has reviewed behavioural and physiological modalities with the strongest association with pre‐clinical disease.ResultOver 140 modalities were identified from the review and were shortlisted to create the version 1 digital toolkit. This first version includes Mezurio and Longevity smartphone apps, a Fitbit charge 4 activity tracker and Dreem 3 sleep headband. This Toolkit was further refined through patient and public involvement studies and collects 26 measures related to 7 aspects of behaviour and physiology (cognition, neural activity, physical activity, heart rate, fine motor movement, sleep, language and speech). The Toolkit is now being used to collect digital data in four international cohorts (Boston University Alzheimer’s Disease Research Center ‐ BU ADRC; The predictors of COgnitive DECline in attenders of memory clinic using digital devices ‐ CODEC‐2; Western Australia Memory Study ‐ WAMS; Healthy Brain Aging ‐ HBA), alongside prospective and retrospective clinical data, to inform the development of machine learning models.ConclusionEDoN will build models with digital markers, validating them against other biomarkers to predict dementia subtypes and individualised disease trajectories. Based on the outputs of the initial models, EDoN will go through a series of iterations of cohort engagement, modality and tool refinement, and data collection. Workstreams are underway to inform data security, privacy, ethics and open policy research, as well as considering the integration of the final EDoN Toolkit into healthcare systems globally. EDoN aims to deliver a cost‐effective, low burden and population‐wide method for early detection of dementia‐causing diseases that will benefit the public, patients, carers, researchers and clinicians, as well as the broader healthcare system and the delivery of new therapies.

Energy Consumption of Electric Vehicles: Analysis of Selected Parameters Based on Created Database

Electric vehicles in a short time will make up the majority of the fleet of vehicles used in general. This state of affairs will generate huge sets of data, which can be further investigated. The paper presents a methodology for the analysis of electric vehicle data, with particular emphasis on the energy consumption parameter. The prepared database contains data for 123 electric vehicles for analysis. Data analysis was carried out in a Python environment with the use of the dabl API library. Presentation of the results was made on the basis of data classification for continuous and categorical features vs. target parameters. Additionally, a heatmap Pearson correlation coefficient was performed to correlate the energy consumption parameter with the other parameters studied. Through the data classification for the studied dataset, it can be concluded that there is no correlation against energy consumption for the parameter charging speed; in contrast, for the parameters range and maximum velocity, a positive correlation can be observed. The negative correlation with the parameter energy consumption is for the parameter acceleration to 100 km/h. The methodology presented to assess data from electric vehicles can be scalable for another dataset to prepare data for creating machine learning models, for example.

Dissociating COVID-19 from other respiratory infections based on acoustic, motor coordination, and phonemic patterns

In the face of the global pandemic caused by the disease COVID-19, researchers have increasingly turned to simple measures to detect and monitor the presence of the disease in individuals at home. We sought to determine if measures of neuromotor coordination, derived from acoustic time series, as well as phoneme-based and standard acoustic features extracted from recordings of simple speech tasks could aid in detecting the presence of COVID-19. We further hypothesized that these features would aid in characterizing the effect of COVID-19 on speech production systems. A protocol, consisting of a variety of speech tasks, was administered to 12 individuals with COVID-19 and 15 individuals with other viral infections at University Hospital Galway. From these recordings, we extracted a set of acoustic time series representative of speech production subsystems, as well as their univariate statistics. The time series were further utilized to derive correlation-based features, a proxy for speech production motor coordination. We additionally extracted phoneme-based features. These features were used to create machine learning models to distinguish between the COVID-19 positive and other viral infection groups, with respiratory- and laryngeal-based features resulting in the highest performance. Coordination-based features derived from harmonic-to-noise ratio time series from read speech discriminated between the two groups with an area under the ROC curve (AUC) of 0.94. A longitudinal case study of two subjects, one from each group, revealed differences in laryngeal based acoustic features, consistent with observed physiological differences between the two groups. The results from this analysis highlight the promise of using nonintrusive sensing through simple speech recordings for early warning and tracking of COVID-19.

Data Augmentation and Random Multi-Model Deep Learning for Data Classification

In the machine learning (ML) paradigm, data augmentation serves as a regularization approach for creating ML models. The increase in the diversification of training samples increases the generalization capabilities, which enhances the prediction performance of classifiers when tested on unseen examples. Deep learning (DL) models have a lot of parameters, and they frequently overfit. Effectively, to avoid overfitting, data plays a major role to augment the latest improvements in DL. Nevertheless, reliable data collection is a major limiting factor. Frequently, this problem is undertaken by combining augmentation of data, transfer learning, dropout, and methods of normalization in batches. In this paper, we introduce the application of data augmentation in the field of image classification using Random Multi-model Deep Learning (RMDL) which uses the association approaches of multiDL to yield random models for classification. We present a methodology for using Generative Adversarial Networks (GANs) to generate images for data augmenting. Through experiments, we discover that samples generated by GANs when fed into RMDL improve both accuracy and model efficiency. Experimenting across both MNIST and CIAFAR-10 datasets show that, error rate with proposed approach has been decreased with different random models.