Linear Learning Model Research Articles

Prediction of seven-day compressive strength of field concrete

This study has explored nine machine learning methods that cover linear, non-linear, and ensemble learning models to predict the compressive strength of field concrete at 7 days. Seven concrete constituents (water, cement, coarse aggregate, fine aggregate, fly ash, superplasticizing admixture, and water reducing admixture) are used as the predictors. A dataset of 12,107 field concrete observations associated with 25 unique mix designs has been used to train and test the predictive models. Evaluated against seven performance metrics, it is found that non-linear models perform better than linear models in general and that the random forest model of ensemble learning performs the best. Compared to previous studies, the models of this study significantly improve in terms of the various performance metrics. Besides, this study confirms that data visualization is useful in learning about and summarizing the data, understanding the relationships of the variables, and making pre-modeling assumptions. This study has also assessed the relative importance of the seven concrete constituents (an aspect previous studies had not investigated), and identified cement, water reducing admixture and fine aggregate as the top three most significant constituents in the development of the seven-day compressive strength.

Heart Failure Prediction with Machine Learning: A Comparative Study

Heart failure is a worldwide healthy problem affecting more than 550,000 people every year. A better prediction for this disease is one of the key approaches of decreasing its impact. Both linear and machine learning models are used to predict heart failure based on various data as inputs, e.g., clinical features. In this paper, we give a comparative study of 18 popular machine learning models for heart failure prediction, with z-score or min-max normalization methods and Synthetic Minority Oversampling Technique (SMOTE) for the imbalance class problem which is often seen in this problem. Our results demonstrate the superiority of using z-score normalization and SMOTE for heart failure prediction.

Machine learning algorithm improves accuracy of ortho-K lens fitting in vision shaping treatment

PurposeTo construct a machine learning (ML)-based model for estimating the alignment curve (AC) curvature in orthokeratology lens fitting for vision shaping treatment (VST), which can minimize the number of lens trials, improving efficiency while maintaining accuracy, with regards to its improvement over a previous calculation method. MethodsData were retrospectively collected from the clinical case files of 1271 myopic subjects (1271 right eyes). The AC curvatures calculated with a previously published algorithm were used as the target data sets. Four kinds of machine learning algorithms were implemented in the experimental analyses to predict the targeted AC curvatures: robust linear regression models, support vector machine (SVM) regression models with linear kernel functions, bagging decision trees, and Gaussian processes. The previously published calculation method and the novel machine learning method were then compared to assess the final parameters of ordered lenses. ResultsThe linear SVM and Gaussian process machine learning models achieved the best performance. The input variables included sex, age, horizontal visible iris diameter (HVID), spherical refraction (SER), cylindrical refraction, eccentricity value (e value), flat K (K1) and steep K (K2) readings, anterior chamber depth (ACD), and axial length (AL). The R-squared values for the output AC1K1, AC1K2 and AC2K1 values were 0.91, 0.84, and 0.73, respectively. The previous calculation method and machine learning methods displayed excellent consistency, and the proposed methods performed best based on flat K reading and e values. ConclusionsThe ML model can provide practitioners with an efficient method for estimating the AC curvatures of VST lenses and reducing the probability of cross-infection originating from trial lenses, which is especially useful during pandemics, such as that for COVID-19.

A machine learning approach to forecasting carry trade returns

ABSTRACT Carry trade refers to a risky arbitrage in interest rate differentials between two currencies. Persistent excess carry trade returns pose a challenge to foreign exchange market efficiency. Using a data set of 10 currencies between 1990 and 2017, we find (i) a machine learning model, long short-term memory (LSTM) networks, forecast carry trade returns better than linear and threshold models and other machine learning models; and (ii) excess carry trade returns deteriorate after the 2007–2008 global financial crisis in all model forecasts, indicating that the uncovered interest rate parity may still hold in the long run.

Detection of mind wandering using EEG: Within and across individuals.

Mind wandering is often characterized by attention oriented away from an external task towards our internal, self-generated thoughts. This universal phenomenon has been linked to numerous disruptive functional outcomes, including performance errors and negative affect. Despite its prevalence and impact, studies to date have yet to identify robust behavioral signatures, making unobtrusive, yet reliable detection of mind wandering a difficult but important task for future applications. Here we examined whether electrophysiological measures can be used in machine learning models to accurately predict mind wandering states. We recorded scalp EEG from participants as they performed an auditory target detection task and self-reported whether they were on task or mind wandering. We successfully classified attention states both within (person-dependent) and across (person-independent) individuals using event-related potential (ERP) measures. Non-linear and linear machine learning models detected mind wandering above-chance within subjects: support vector machine (AUC = 0.715) and logistic regression (AUC = 0.635). Importantly, these models also generalized across subjects: support vector machine (AUC = 0.613) and logistic regression (AUC = 0.609), suggesting we can reliably predict a given individual’s attention state based on ERP patterns observed in the group. This study is the first to demonstrate that machine learning models can generalize to “never-seen-before” individuals using electrophysiological measures, highlighting their potential for real-time prediction of covert attention states.

Treatment with a neutrophil elastase inhibitor and ofloxacin reduces P. aeruginosa burden in a mouse model of chronic suppurative otitis media

Chronic suppurative otitis media (CSOM) is a widespread, debilitating problem with poorly understood immunology. Here, we assess the host response to middle ear infection over the course of a month post-infection in a mouse model of CSOM and in human subjects with the disease. Using multiparameter flow cytometry and a binomial generalized linear machine learning model, we identified Ly6G, a surface marker of mature neutrophils, as the most informative factor of host response driving disease in the CSOM mouse model. Consistent with this, neutrophils were the most abundant cell type in infected mice and Ly6G expression tracked with the course of infection. Moreover, neutrophil-specific immunomodulatory treatment using the neutrophil elastase inhibitor GW 311616A significantly reduces bacterial burden relative to ofloxacin-only treated animals in this model. The levels of dsDNA in middle ear effusion samples are elevated in both humans and mice with CSOM and decreased during treatment, suggesting that dsDNA may serve as a molecular biomarker of treatment response. Together these data strongly implicate neutrophils in the ineffective immune response to P. aeruginosa infection in CSOM and suggest that immunomodulatory strategies may benefit drug-tolerant infections for chronic biofilm-mediated disease.

Structural signatures for thermodynamic stability in vitreous silica: Insight from machine learning and molecular dynamics simulations

The structure-thermodynamic stability relationship in vitreous silica is investigated using machine learning and a library of 24,157 inherent structures generated from melt-quenching and replica exchange molecular dynamics simulations. We find the thermodynamic stability, i.e., enthalpy of the inherent structure ($e_{\mathrm{IS}}$), can be accurately predicted by both linear and nonlinear machine learning models from numeric structural descriptors commonly used to characterize disordered structures. We find short-range features become less indicative of thermodynamic stability below the fragile-to-strong transition. On the other hand, medium-range features, especially those between 2.8-~6 $\unicode{x212B}$;, show consistent correlations with $e_{\mathrm{IS}}$ across the liquid and glass regions, and are found to be the most critical to stability prediction among features from different length scales. Based on the machine learning models, a set of five structural features that are the most predictive of the silica glass stability is identified.

Comparison of linear, penalized linear and machine learning models predicting hospital visit costs from chronic disease in Thailand

Generally, health care costs from chronic diseases have positive skew and this gives problems on using traditional statistical models. Machine learning is a conventional method producing accurate prediction with large sample size. However, much of the comparison performance between statistical methods and machine learning for such data remains scattered. This study aimed to compare linear, penalized linear and machine learning models for their prediction performance of hospital visit costs from chronic disease, in Thailand. A total of 18,342 hospital visit records were obtained from Suratthani tertiary hospital in southern Thailand, which contained data from 2016 on chronic patients of Diagnosis-Related Groups (DRGs). The prediction performance on hospital visit costs by linear, penalized linear and machine learning models were compared using both original dataset and datasets expanded in size two- and four-fold by using bootstrap. The mean age of patients was 56.3 ± 22.6 years with 55.6% of visits by males. The median hospital cost was 16,662 Baht per visit. The random forest (RF) model had the best predictive performance of hospital visit costs for all sizes of dataset with the smallest prediction errors, whereas ridge linear regression had the poorest prediction performance with the largest prediction errors. Machine learning models had better prediction performance with enlarged sample sizes whereas linear and penalized linear models did not. On modeling big data for prediction, machine learning models are preferable, whereas linear and penalized linear models' predictions are not affected by increasing the sample size.

Robust Subject-Independent P300 Waveform Classification via Signal Pre-Processing and Deep Learning

Brain Computer Interfaces (BCIs) are capable of processing neural stimuli using electroencephalogram (EEG) measurements to aid communication capabilities. Yet, BCIs often require extensive calibration steps in order to be tuned to specific users. In this work, we develop a subject independent P300 classification framework, which eliminates the need for user-specific calibration. We begin by employing a series of pre-processing steps, where, among other steps, we consider different trial averaging methodologies and various EEG electrode configurations. We then consider three distinct deep learning architectures and two linear machine learning models as P300 signal classifiers. Through evaluation on three datasets, and in comparison to three benchmark P300 classification frameworks, we find that averaging up to seven trials while using eight specific electrode channels on a two-layered convolutional neural network (CNN) leads to robust subject independent P300 classification. In this capacity, our method achieves greater than a 0.20 gain in AUC in comparison to prior P300 classification methods. In addition, our proposed framework is computationally efficient with training time gains of greater than 3×, compared to linear machine learning models, and online evaluation time speedups of up to 2× compared to benchmark methods.

Long Term Solar Power Generation Prediction using Adaboost as a Hybrid of Linear and Non-linear Machine Learning Model

The usage of renewable energy sources has increased manifolds in terms of electric utilities. From other non-conventional sources, solar energy has been seen as a promising and convenient source used around the globe. In terms of meeting the daily requirements, solar energy has huge potential to fulfil the world’s demand. However, firstly the characteristic of solar energy is very unpredictable and intermittent due to variation of weather. Secondly, the optimization and the planning of smart grid effect the operation of PV system. Thus, prediction on the long horizon is needed to address this problem. Nevertheless, long term forecasting of solar power generation is deliberated as a challenging problem. Therefore, this paper proposes a 10 day ahead solar power forecasting using combination of linear and non-linear machine learning models. At first, the outputs are generated from Recurrent Neural Network (RNN), Support Vector Machine (SVM) and Autoregressive with exogenous variable (ARX). Later on, these three outputs are combined and are made as a strong classifier with the Adaptive boost (Adaboost) algorithm. The simulations were conducted on the data obtained from real PV plant. By the experimental results and discussion, it was endogenously concluded that the combination of all techniques with the Adaboost have increased the performances and showing the high accuracy as compare to the individual machine learning models. The hybrid Adaboost shows %MAPE 8.88, which proven high accuracy. While on the other hand, for the individual technique, RNN shows 10.88, SVM reveals 11.78 and ARX gives 13.00 of percentage MAPE. The improvement proves that combination of techniques performs better than individual models and proclaims the high accuracy.

Predicting Risk of Mortality in Pediatric ICU Based on Ensemble Step-Wise Feature Selection.

Background. Prediction of mortality risk in intensive care units (ICU) is an important task. Data-driven methods such as scoring systems, machine learning methods, and deep learning methods have been investigated for a long time. However, few data-driven methods are specially developed for pediatric ICU. In this paper, we aim to amend this gap-build a simple yet effective linear machine learning model from a number of hand-crafted features for mortality prediction in pediatric ICU.Methods. We use a recently released publicly available pediatric ICU dataset named pediatric intensive care (PIC) from Children's Hospital of Zhejiang University School of Medicine in China. Unlike previous sophisticated machine learning methods, we want our method to keep simple that can be easily understood by clinical staffs. Thus, an ensemble step-wise feature ranking and selection method is proposed to select a small subset of effective features from the entire feature set. A logistic regression classifier is built upon selected features for mortality prediction.Results. The final predictive linear model with 11 features achieves a 0.7531 ROC-AUC score on the hold-out test set, which is comparable with a logistic regression classifier using all 397 features (0.7610 ROC-AUC score) and is higher than the existing well known pediatric mortality risk scorer PRISM III (0.6895 ROC-AUC score).Conclusions. Our method improves feature ranking and selection by utilizing an ensemble method while keeping a simple linear form of the predictive model and therefore achieves better generalizability and performance on mortality prediction in pediatric ICU.

A study of computational design optimization of concrete mixes

This paper presents a thorough overview of optimization research on the design and proportioning of concrete mixes. The necessity for designers and decision-makers to proportion concrete mixes that meet various sometimes conflicting performance criteria, including as cost, workability, mechanical characteristics, durability, and environmental sustainability, is driving mixture design optimization. We begin with a discussion of typical mathematical problem formulations, choices, goals, and constraints related to concrete mixture design optimization in this study. Following that, we look at the many kinds of models used to simulate concrete characteristics, such as a range of linear combination, statistical, machine learning, and physics-based models that are needed to optimize a mixture's proportions. Then, in the context of the material reviewed, we examine and discuss computer techniques for optimizing concrete mixes. Finally, in light of the present state of the art, we highlight and analyze current trends and possibilities for improving the area of concrete mixture design optimization.

Interaction modeling and classification scheme for augmenting the response accuracy of human-robot interaction systems.

Human-robot interaction (HRI) is becoming a current research field for providing granular real-time applications and services through physical observation. Robotic systems are designed to handle the roles of humans and assist them through intrinsic sensing and commutative interactions. These systems handle inputs from multiple sources, process them, and deliver reliable responses to the users without delay. Input analysis and processing is the prime concern for the robotic systems to understand and resolve the queries of the users. In this manuscript, the Interaction Modeling and Classification Scheme (IMCS) is introduced to improve the accuracy of HRI. This scheme consists of two phases, namely error classification and input mapping. In the error classification process, the input is analyzed for its events and conditional discrepancies to assign appropriate responses in the input mapping phase. The joint process is aided by a linear learning model to analyze the different conditions in the event and input detection. The performance of the proposed scheme shows that it is capable of improving the interaction accuracy by reducing the ratio of errors and interaction response by leveraging the information extraction from the discrete and successive human inputs. The fetched data are analyzed by classifying the errors at the initial stage to achieve reliable responses.

APPLICATION OF MACHINE LEARNING MODELS IN ENROLLMENT AND STUDENT TRAINING AT VIETNAMESE UNIVERSITIES

In Vietnam, since 2015, the Ministry of Education and Training of Vietnam has decided to abolish university entrance exams and advocates the use of high school graduation exam results of candidates for admission to go to universities. The 2015 and 2016 exam questions for the Math exam are the essay questions. From 2017 up to now, the Ministry of Education and Training of Viet nam has applied the form of multiple-choice exams for Mathematics in the high school graduation exam. There are many mixed opinions about the impact of this form of examination and admission on the quality of university students. In particular, the switch from the form of essay examination to multiple-choice exams led the entire Vietnam Mathematical Association at that time to send recommendations on continuing to maintain the form of essay examination for mathematics. The purposes of this article are analysisand evaluation the effects of relevant factors on the academic performance of advanced math students of university students, and offer solutions to optimize university entrance exam. The data set was provided by Training Management Department and Training Quality Control and Testing Laboratory of the University of Finance – Marketing. This dataset includes information about math high school graduation test scores, learning process scores (scores assessed by direct instructors), and advanced math course end test scores of 2834 students in courses from 2015 to 2019. Linear and non-linear regression machine learning models were used to solve the tasks given in this article. An analysis of the data was conducted to reveal the advantages and disadvantages of the change in university enrollment of the Vietnamese Ministry of Education and Training. Tools from the Python libraries have been supported and used effectively in the process of solving problems. Through building and surveying the model, there are suggestions and solutions to problems in enrollment and input quality assurance. Specifically, in the preparation of entrance exams, the entrance exam questions should not exceed 61-66 % of multiple choice questions.

Evaluation of Linear and Machine Learning Models for Determining Pedotransfer Functions

Modelling of water content and the transport of water in soil has become a useful tool in simulating agricultural productivity or in solving various hydrological analyses. For instance, optimum irrigation management requires a systematic estimation of the soil-moisture to determine both the appropriate amount and timing of irrigation. Soil characteristics appear as an essential input in the numerical simulation of a soil-water regime. A critical physical property used in the description of a soil-water regime in such modelling is a soil water retention curve. This paper aims to evaluate so-called pedotransfer functions, which helps to assess the soil water retention curve easier than by standard complex and lengthy procedure involving both field and laboratory work. As a case study Zahorska Lowland, which is located in central Europe in the western part of Slovakia, was selected. The frequent occurrence of dry years in this area results in the necessity to construct irrigation systems in this area, so modelling water content in the soil is an important task here. This study aims to support such modelling with determining pedotransfer functions. Authors compare linear methods (multiple linear regression, LASSO regularized regression) and CatBoost machine learning model.

Asynchrony Between Individual and Government Actions Accounts for Disproportionate Impact of COVID-19 on Vulnerable Communities

IntroductionPreviously estimated effects of social distancing do not account for changes in individual behavior before the implementation of stay-at-home policies or model this behavior in relation to the burden of disease. This study aims to assess the asynchrony between individual behavior and government stay-at-home orders, quantify the true impact of social distancing using mobility data, and explore the sociodemographic variables linked to variation in social distancing practices.MethodsThis study was a retrospective investigation that leveraged mobility data to quantify the time to behavioral change in relation to the initial presence of COVID-19 and the implementation of government stay-at-home orders. The impact of social distancing that accounts for both individual behavior and testing data was calculated using generalized mixed models. The role of sociodemographics in accounting for variation in social distancing behavior was modeled using a 10-fold cross-validated elastic net (linear machine learning model). Analysis was conducted in April‒July 2020.ResultsAcross all the 1,124 counties included in this analysis, individuals began to socially distance at a median of 5 days (IQR=3−8) after 10 cumulative cases of COVID-19 were confirmed in their state, with state governments taking a median of 15 days (IQR=12−19) to enact stay-at-home orders. Overall, people began social distancing at a median of 12 days (IQR=8−17) before their state enacted stay-at-home orders. Of the 16 studies included in the review, 13 exclusively used government dates as a proxy for social distancing behavior, and none accounted for both testing and mobility. Using government stay-at-home dates as a proxy for social distancing (10.2% decrease in the number of daily cases) accounted for only 55% of the true impact of the intervention when compared with estimates using mobility (18.6% reduction). Using 10-fold cross-validation, 23 of 43 sociodemographic variables were significantly and independently predictive of variation in individual social distancing, with delays corresponding to an increase in a county's proportion of people without a high school diploma and proportion of racial and ethnic minorities.ConclusionsThis retrospective analysis of mobility patterns found that social distancing behavior occurred well before the onset of government stay-at-home dates. This asynchrony leads to the underestimation of the impact of social distancing. Sociodemographic characteristics associated with delays in social distancing can help explain the disproportionate case burden and mortality among vulnerable communities.

Quantum algorithm for the nonlinear dimensionality reduction with arbitrary kernel

Dimensionality reduction (DR) techniques play an extremely critical role in the data mining and pattern recognition field. However, most DR approaches involve large-scale matrix computations, which cause too high running complexity to implement in the big data scenario efficiently. The recent developments in quantum information processing provide a novel path to alleviate this problem, where a potential quantum acceleration can be obtained comparing with the classical counterpart. Nevertheless, existing proposals for quantum DR methods faced the common dilemma of the nonlinear generalization owing to the intrinsic linear limitation of quantum computation. In this paper, an architecture to simulate the arbitrary nonlinear kernels on a universal quantum computer is illustrated and further propose the quantum kernel principal component analysis (QKPCA) algorithm. The key idea is employing the truncated Taylor expansion to approximate the arbitrary nonlinear kernel within the fixed error and then constructing the corresponding Hamiltonian simulation for the quantum phase estimation algorithm. It is demonstrated theoretically that the QKPCA is qualified for the nonlinear DR task while the exponential speedup is also maintained. In addition, this research has the potential ability to develop other quantum DR approaches and existing linear quantum machine learning models.

Identification of an epigenetic signature in human induced pluripotent stem cells using a linear machine learning model

The use of human induced pluripotent stem cells (iPSCs), used as an alternative to human embryonic stem cells (ESCs), is a potential solution to challenges, such as immune rejection, and does not involve the ethical issues concerning the use of ESCs in regenerative medicine, thereby enabling developments in biological research. However, comparative analyses from previous studies have not indicated any specific feature that distinguishes iPSCs from ESCs. Therefore, in this study, we established a linear classification-based learning model to distinguish among ESCs, iPSCs, embryonal carcinoma cells (ECCs), and somatic cells on the basis of their DNA methylation profiles. The highest accuracy achieved by the learned models in identifying the cell type was 94.23%. In addition, the epigenetic signature of iPSCs, which is distinct from that of ESCs, was identified by component analysis of the learned models. The iPSC-specific regions with methylation fluctuations were abundant on chromosomes 7, 8, 12, and 22. The method developed in this study can be utilized with comprehensive data and widely applied to many aspects of molecular biology research.

Sentiment Analysis on Amazon Fine Food Reviews by using Linear Machine Learning Models

Sentiment Analysis on Amazon Fine Food Reviews by using Linear Machine Learning Models

Using silicon photovoltaic cells and machine learning and neural network algorithms for visible-light positioning systems

We propose and experimentally demonstrate visible-light positioning (VLP) systems using silicon photovoltaic cells (Si-PVCs) and machine learning and neural network algorithms. Both angle-of-arrival (AOA)-based and received-signal-strength (RSS)-based VLP systems are evaluated and compared. The Si-PVC could also provide energy harvesting to store received optical power for the mobile unit. Here, second-order linear regression machine learning (RML) model and two-layer neural network are implemented in both AOA-based and RSS-based VLP systems to enhance the positioning accuracy. The root mean square (RMS) average positioning error of the AOA-based VLP system is reduced from 7.22 to 3.46 and to 2.99 cm when using the RML and neural network, respectively. The RMS average positioning error of the RSS-based VLP system is reduced from 7.07 to 3.01 and to 2.60 cm when using the RML and neural network, respectively. The experimental results clearly illustrate that the proposed schemes can significantly improve the positioning accuracy.