Machine Learning-based Strategy Research Articles

Community screening for dementia among older adults in China: a machine learning-based strategy

BackgroundDementia is a leading cause of disability in people older than 65 years worldwide. However, diagnosing dementia in its earliest symptomatic stages remains challenging. This study combined specific questions from the AD8 scale with comprehensive health-related characteristics, and used machine learning (ML) to construct diagnostic models of cognitive impairment (CI).MethodsThe study was based on the Shenzhen Healthy Ageing Research (SHARE) project, and we recruited 823 participants aged 65 years and older, who completed a comprehensive health assessment and cognitive function assessments. Permutation importance was used to select features. Five ML models using BalanceCascade were applied to predict CI: a support vector machine (SVM), multilayer perceptron (MLP), AdaBoost, gradient boosting decision tree (GBDT), and logistic regression (LR). An AD8 score ≥ 2 was used to define CI as a baseline. SHapley Additive exPlanations (SHAP) values were used to interpret the results of ML models.ResultsThe first and sixth items of AD8, platelets, waist circumference, body mass index, carcinoembryonic antigens, age, serum uric acid, white blood cells, abnormal electrocardiogram, heart rate, and sex were selected as predictive features. Compared to the baseline (AUC = 0.65), the MLP showed the highest performance (AUC: 0.83 ± 0.04), followed by AdaBoost (AUC: 0.80 ± 0.04), SVM (AUC: 0.78 ± 0.04), GBDT (0.76 ± 0.04). Furthermore, the accuracy, sensitivity and specificity of four ML models were higher than the baseline. SHAP summary plots based on MLP showed the most influential feature on model decision for positive CI prediction was female sex, followed by older age and lower waist circumference.ConclusionsThe diagnostic models of CI applying ML, especially the MLP, were substantially more effective than the traditional AD8 scale with a score of ≥ 2 points. Our findings may provide new ideas for community dementia screening and to promote such screening while minimizing medical and health resources.

Carbon Emission Optimization of Ultra-High-Performance Concrete Using Machine Learning Methods.

Due to its exceptional qualities, ultra-high-performance concrete (UHPC) has recently become one of the hottest research areas, although the material's significant carbon emissions go against the current development trend. In order to lower the carbon emissions of UHPC, this study suggests a machine learning-based strategy for optimizing the mix proportion of UHPC. To accomplish this, an artificial neural network (ANN) is initially applied to develop a prediction model for the compressive strength and slump flow of UHPC. Then, a genetic algorithm (GA) is employed to reduce the carbon emissions of UHPC while taking into account the strength, slump flow, component content, component proportion, and absolute volume of UHPC as constraint conditions. The outcome is then supported by the results of the experiments. In comparison to the experimental results, the research findings show that the ANN model has excellent prediction accuracy with an error of less than 10%. The carbon emissions of UHPC are decreased to 688 kg/m3 after GA optimization, and the effect of optimization is substantial. The machine learning (ML) model can provide theoretical support for the optimization of various aspects of UHPC.

Assessing risk factors for malnutrition among women in Bangladesh and forecasting malnutrition using machine learning approaches

BackgroundThis paper presents an in-depth examination of malnutrition in women in Bangladesh. Malnutrition in women is a major public health issue related to different diseases and has negative repercussions for children, such as premature birth, decreased infection resistance, and an increased risk of death. Moreover, malnutrition is a severe problem in Bangladesh. Data from the Bangladesh Demographic Health Survey (BDHS) conducted in 2017-18 was used to identify risk factors for malnourished women and to create a machine learning-based strategy to detect their nutritional status.MethodsA total of 17022 women participants are taken to conduct the research. All the participants are from different regions and different ages. A chi-square test with a five percent significance level is used to identify possible risk variables for malnutrition in women and six machine learning-based classifiers (Naïve Bayes, two types of Decision Tree, Logistic Regression, Random Forest, and Gradient Boosting Machine) were used to predict the malnutrition of women. The models are being evaluated using different parameters like accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$F_1$$\\end{document} score, and area under the curve (AUC).ResultsDescriptive data showed that 45% of the population studied were malnourished women, and the chi-square test illustrated that all fourteen variables are significantly associated with malnutrition in women and among them, age and wealth index had the most influence on their nutritional status, while water source had the least impact. Random Forest had an accuracy of 60% and 60.2% for training and test data sets, respectively. CART and Gradient Boosting Machine also had close accuracy like Random Forest but based on other performance metrics such as kappa and F1\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$F_1$$\\end{document} scores Random Forest got the highest rank among others. Also, it had the highest accuracy and F1\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$F_1$$\\end{document} scores in k-fold validation along with the highest AUC (0.604).ConclusionThe Random Forest (RF) approach is a reasonably superior machine learning-based algorithm for forecasting women’s nutritional status in Bangladesh in comparison to other ML algorithms investigated in this work. The suggested approach will aid in forecasting which women are at high susceptibility to malnutrition, hence decreasing the strain on the healthcare system.

De Novo Generation and Identification of Novel Compounds with Drug Efficacy Based on Machine Learning.

One of the main challenges in small molecule drug discovery is finding novel chemical compounds with desirable activity. Traditional drug development typically begins with target selection, but the correlation between targets and disease remains to be further investigated, and drugs designed based on targets may not always have the desired drug efficacy. The emergence of machine learning provides a powerful tool to overcome the challenge. Herein, a machine learning-based strategy is developed for de novo generation of novel compounds with drug efficacy termed DTLS (Deep Transfer Learning-based Strategy) by using dataset of disease-direct-related activity as input. DTLS is applied in two kinds of disease: colorectal cancer (CRC) and Alzheimer's disease (AD). In each case, novel compound is discovered and identified in in vitro and in vivo disease models. Their mechanism of actionis further explored. The experimental results reveal that DTLS can not only realize the generation and identification of novel compounds with drug efficacy but also has the advantage of identifying compounds by focusing on protein targets to facilitate the mechanism study. This work highlights the significant impact of machine learning on the design of novel compounds with drug efficacy, which provides a powerful new approach to drug discovery.

DDoS attack detection in smart grid network using reconstructive machine learning models.

Network attacks pose a significant challenge for smart grid networks, mainly due to the existence of several multi-directional communication devices coupling consumers to the grid. One of the network attacks that can affect the smart grid is the distributed denial of service (DDoS), where numerous compromised communication devices/nodes of the grid flood the smart grid network with false data and requests, leading to disruptions in smart meters, data servers, and the state estimator, ultimately effecting the services for end-users. Machine learning-based strategies show distinctive benefits in resolving the challenge of securing the network from DDoS attacks. Regardless, a notable hindrance in deploying machine learning-based techniques is the requirement of model retraining whenever new attack classes arise. Practically, disrupting the normal operations of smart grid is really discouraged. To handle this challenge effectively and detect DDoS attacks without major disruptions, we propose the deployment of reconstructive deep learning techniques. A primary benefit of our proposed technique is the minimum disruption during the introduction of a new attack class, even after complete deployment. We trained several deep and shallow reconstructive models to get representations for each attack type separately, and we performed attack detection by class-specific reconstruction error-based classification. Our technique experienced rigid evaluation via multiple experiments using two well-acknowledged standard databases exclusively for DDoS attacks, including their subsets. Later, we performed a comparative estimation of our outcomes against six methods prevalent within the same domain. Our outcomes reveal that our technique attained higher accuracy, and notably eliminates the requirement of a complete model retraining in the event of the introduction of new attack classes. This method will not only boost the security of smart grid networks but also ensure the stability and reliability of normal operations, protecting the critical infrastructure from ever-evolving network attacks. As smart grid is advancing rapidly, our approach proposes a robust and adaptive way to overcome the continuous challenges posed by network attacks.

Development of prognostic model and multivariate analysis for breast cancer survival patients using SEER database

Background: Many studies employed machine learning (ML) to forecast the prognosis of breast cancer (BC) patients and discovered that the ML model showed high individualized forecasting ability. Breast cancer is the most frequent kind of carcinoma in women globally and ranks as the leading cause of death in women. Objectives: This study intends to use the Surveillance, Epidemiology, and End Results dataset to categorize breast carcinoma cases’ alive and dead conditions. Deep learning and machine learning have been extensively utilized in clinical studies to address various categorization problems due to their ability to manage massive data sets in an organized manner. Pre-processing the data allows it to be visualized and analyzed for making critical choices. This study describes a realistic machine learning-based strategy for categorizing the SEER breast cancer dataset. Materials and methods: We employed classification and machine learning algorithms to classify breast cancer mortality. Four well-known classification ML algorithms were employed in this study. To identify risk factors, we employed multivariate analysis using the data set. Results: The decision tree performed the best accuracy (0.914) among all the models. T4 stage (β=1.4, p<0.001, OR=4.22, 95% CI (2.06-8.64), N2 stage (β=0.39, p=0.008, OR= 1.49, 95% CI (1.111-1.997) found to be major risk factors for breast cancer mortality using multivariate analysis. Conclusion: The significant prognostic variables affecting the breast carcinoma survival rates reported in the current research are relevant and might be turned into decision support systems in the medical realm.

Capturing Directivity in Probabilistic Seismic Hazard Analysis for New Zealand: Challenges, Implications, and a Machine Learning Approach for Implementation

ABSTRACT The proximity of fast-slipping crustal faults to urban areas may result in pulse-like ground motions from rupture directivity, which can contribute to increased levels of damage even for engineered structures. Systematic modeling of directivity within probabilistic seismic hazard analysis (PSHA) remains challenging to implement at the regional scale, despite the availability of directivity models in the literature. In the process of developing the 2022 National Seismic Hazard Model for New Zealand (2022 NSHM), we explored the feasibility and impact of modeling directivity for PSHA at a national scale using the previous generation 2010 NSHM. The results of this analysis allowed us to quantify the impact of directivity on the resulting seismic hazard maps for New Zealand and gain insights into the factors that contribute to the expected increases (and decreases) in ground-motion level. For the 2022 NSHM, the earthquake rupture forecast (ERF) seismogenic source models introduced enormous challenges for directivity modeling due to the abundance of large multisegment or multifault ruptures with complex geometries. To overcome these challenges, we applied a machine learning-based strategy to “overfit” an artificial neural network to capture the distributions of directivity amplification and its variability for each unique rupture in the earthquake rupture forecast. This produces a compact representation of the spatial fields of amplification that are computationally efficient to generate within a complete PSHA calculation for the 2022 NSHM. This flexible and reproducible framework facilitates the implementation of directivity in PSHA at a regional scale for complex ERF source models and opens up the possibility of more complex characterization of epistemic uncertainties for near-source ground motion in practice.

Illegal, unreported and unregulated fishing detection with machine learning

Illegal, unreported and unregulated fishing is a worldwide problem that causes local and global economic losses, depletes natural resources, alters our diverse ecosystems and takes an undue toll on fisheries. This study describes a machine learning-based strategy for response generation. Identifying data storage and processes has led to the initial development of a viable IUU fishing detection system that classifies vessels for IUU fishing by combining (1) the likelihood that a vessel is fishing using geospatially referenced signal data, and (2) whether or not it is classified The likelihood of fish activity is scored for ships to be within the area of interest, and (3) classification of whether the vessel is allowed to enter its habitable area of interest. In this paper, certain parts of the system were prototyped, including using logistic regression to develop highly predictive catch or no-fish classification models for longlines and trawlers, and identifying whether a vessel was within an area of interest process. In addition, many fishing vessel registries have been identified, which regulate the rights of specific vessels to fish in regulated areas of interest. The accuracy with which fishing models can predict the probability of fishing when vessels have longline or trawl gear is acceptable, and can predict vessels with seine gear, but additional research and analysis are needed. "In ROI" classification models should be extended to score their likelihood of being in ROI instead of outputting true/false judgments. Using machine learning and data analysis skills, the project aims to make further efforts to predict IUU fishing in order to enable law enforcement and ultimately significantly reduce or prevent IUU fishing.

Data-Driven Modeling of Appliance Energy Usage

Due to the transition toward the Internet of Everything (IOE), the prediction of energy consumed by household appliances has become a progressively more difficult topic to model. Even with advancements in data analytics and machine learning, several challenges remain to be addressed. Therefore, providing highly accurate and optimized models has become the primary research goal of many studies. This paper analyzes appliance energy consumption through a variety of machine learning-based strategies. Utilizing data recorded from a single-family home, input variables comprised internal temperatures and humidities, lighting consumption, and outdoor conditions including wind speed, visibility, and pressure. Various models were trained and evaluated: (a) multiple linear regression, (b) support vector regression, (c) random forest, (d) gradient boosting, (e) xgboost, and (f) the extra trees regressor. Both feature engineering and hyperparameter tuning methodologies were applied to not only extend existing features but also create new ones that provided improved model performance across all metrics: root mean square error (RMSE), coefficient of determination (R2), mean absolute error (MAE), and mean absolute percentage error (MAPE). The best model (extra trees) was able to explain 99% of the variance in the training set and 66% in the testing set when using all the predictors. The results were compared with those obtained using a similar methodology. The objective of performing these actions was to show a unique perspective in simulating building performance through data-driven models, identifying how to maximize predictive performance through the use of machine learning-based strategies, as well as understanding the potential benefits of utilizing different models.

A machine learning-based strategy for predicting the mechanical strength of coral reef limestone using X-ray computed tomography

Different sedimentary zones in coral reefs lead to significant anisotropy in the pore structure of coral reef limestone (CRL), making it difficult to study mechanical behaviors. With X-ray computed tomography (CT), 112 CRL samples were utilized for training the support vector machine (SVM)-, random forest (RF)-, and back propagation neural network (BPNN)-based models, respectively. Simultaneously, the machine learning model was embedded into genetic algorithm (GA) for parameter optimization to effectively predict uniaxial compressive strength (UCS) of CRL. Results indicate that the BPNN model with five hidden layers presents the best training effect in the data set of CRL. The SVM-based model shows a tendency to overfitting in the training set and poor generalization ability in the testing set. The RF-based model is suitable for training CRL samples with large data. Analysis of Pearson correlation coefficient matrix and the percentage increment method of performance metrics shows that the dry density, pore structure, and porosity of CRL are strongly correlated to UCS. However, the P-wave velocity is almost uncorrelated to the UCS, which is significantly distinct from the law for homogenous geomaterials. In addition, the pore tensor proposed in this paper can effectively reflect the pore structure of coral framework limestone (CFL) and coral boulder limestone (CBL), realizing the quantitative characterization of the heterogeneity and anisotropy of pore. The pore tensor provides a feasible idea to establish the relationship between pore structure and mechanical behavior of CRL.

A Machine Learning Framework to Improve Rat Clearance Predictions and Inform Physiologically Based Pharmacokinetic Modeling.

During drug discovery and development, achieving appropriate pharmacokinetics is key to establishment of the efficacy and safety of new drugs. Physiologically based pharmacokinetic (PBPK) models integrating in vitro-to-in vivo extrapolation have become an essential in silico tool to achieve this goal. In this context, the most important and probably most challenging pharmacokinetic parameter to estimate is the clearance. Recent work on high-throughput PBPK modeling during drug discovery has shown that a good estimate of the unbound intrinsic clearance (CLint,u,) is the key factor for useful PBPK application. In this work, three different machine learning-based strategies were explored to predict the rat CLint,u as the input into PBPK. Therefore, in vivo and in vitro data was collected for a total of 2639 proprietary compounds. The strategies were compared to the standard in vitro bottom-up approach. Using the well-stirred liver model to back-calculate in vivo CLint,u from in vivo rat clearance and then training a machine learning model on this CLint,u led to more accurate clearance predictions (absolute average fold error (AAFE) 3.1 in temporal cross-validation) than the bottom-up approach (AAFE 3.6-16, depending on the scaling method) and has the advantage that no experimental in vitro data is needed. However, building a machine learning model on the bias between the back-calculated in vivo CLint,u and the bottom-up scaled in vitro CLint,u also performed well. For example, using unbound hepatocyte scaling, adding the bias prediction improved the AAFE in the temporal cross-validation from 16 for bottom-up to 2.9 together with the bias prediction. Similarly, the log Pearson r2 improved from 0.1 to 0.29. Although it would still require in vitro measurement of CLint,u., using unbound scaling for the bottom-up approach, the need for correction of the fu,inc by fu,p data is circumvented. While the above-described ML models were built on all data points available per approach, it is discussed that evaluation comparison across all approaches could only be performed on a subset because ca. 75% of the molecules had missing or unquantifiable measurements of the fraction unbound in plasma or in vitro unbound intrinsic clearance, or they dropped out due to the blood-flow limitation assumed by the well-stirred model. Advantageously, by predicting CLint,u as the input into PBPK, existing workflows can be reused and the prediction of the in vivo clearance and other PK parameters can be improved.

The Interacting Multiple Model Smooth Variable Structure Filter for Trajectory Prediction

An autonomous vehicle would benefit from being able to predict trajectories of other vehicles in its vicinity for improved safety. In order for the self-driving car to plan safe trajectories, paths of nearby vehicles are required to be predicted for risk assessment, decision making, and motion planning. In this study, a trajectory prediction algorithm based on the Interacting Multiple Model (IMM) estimation strategy is proposed to predict paths involving lane-changing, lane-keeping, and turning motion. More specifically, the Interacting Multiple Model estimation technique is used with models defined in curvi-linear coordinates to predict a vehicle’s trajectory based on prior behavioral maneuvers. The road geometry is used to help facilitate behavior identification and prediction. Moreover, the combination of a more recently developed estimation technique known as the Generalized Variable Boundary Layer-Smooth Variable Structure Filter and the Interacting Multiple Model Estimator is applied to track, identify behaviors, and predict trajectories of a vehicle. The performance of this technique is compared with a Kalman Filter based formulation using synthetic and experimental data. This model-based strategy is also compared with machine learning-based strategies for trajectory prediction.

An adaptive semi-supervised deep learning-based framework for the detection of Android malware

Positive developments in smartphone usage have led to an increase in malicious attacks, particularly targeting Android mobile devices. Android has been a primary target for malware exploiting security vulnerabilities due to the presence of critical applications, such as banking applications. Several machine learning-based models for mobile malware detection have been developed recently, but significant research is needed to achieve optimal efficiency and performance. The proliferation of Android devices and the increasing threat of mobile malware have made it imperative to develop effective methods for detecting malicious apps. This study proposes a robust hybrid deep learning-based approach for detecting and predicting Android malware that integrates Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM). It also presents a creative machine learning-based strategy for dealing with unbalanced datasets, which can mislead the training algorithm during classification. The proposed strategy helps to improve method performance and mitigate over- and under-fitting concerns. The proposed model effectively detects Android malware. It extracts both temporal and spatial features from the dataset. A well-known Drebin dataset was used to train and evaluate the efficacy of all creative frameworks regarding the accuracy, sensitivity, MAE, RMSE, and AUC. The empirical finding proclaims the projected hybrid ConvLSTM model achieved remarkable performance with an accuracy of 0.99, a sensitivity of 0.99, and an AUC of 0.99. The proposed model outperforms standard machine learning-based algorithms in detecting malicious apps and provides a promising framework for real-time Android malware detection.

An Evaluation of Machine Learning and Big Data Analytics Performance in Cloud Computing and Computer Vision

Although cloud computing is receiving a lot of attention, security remains a significant barrier to its general adoption. Cloud service users frequently worry about data loss, security risks, and availability issues. Because of the accessibility and openness of the huge volume of data amassed by sensors and the web throughout recent years, computer applications have seen a remarkable change from straightforward data processing to machine learning. Two widely used technologies, Big Data and Cloud computing, are the focus of worry in the IT industry. Enormous data sets are put away, handled, and broke down under the possibility of "Big Data." Then again, cloud computing centres around giving the framework to make such systems conceivable in a period and cash saving way. The objective of the review is to survey the Big Data Analytics and Machine learning ideal models for use in cloud computing and computer vision. The programmed data examination of enormous data sets and the production of models for the wide connections between data are the centre highlights of machine learning (ML). The usefulness of machine learning-based strategies for identifying threats in a cloud computing environment is surveyed and compared in this research.

A Review Article On Enhancing Email Spam Filter’s Accuracy Using Machine Learning

In today’s era, almost everyone is using emails on their daily basis. In our proposed research, we suggest a machine learning-based strategy for enhancing email spam filters' accuracy. Traditional rule-based filters have grown less effective as spam emails have multiplied exponentially. Models can be trained to identify emails as spam or not using machine learning algorithms, particularly supervised learning. We need to create a simple and straightforward machine learning model in order to reach more accurate results while categorizing email spam. We picked the Naive Bayes technique for our model since it is quicker and more accurate than other algorithms. The suggested method can have incorporated into current email systems to enhance spam filtering functionality. This review paper provides an overview of the machine learning model we have suggested.

Sequential virtual screening collaborated with machine-learning strategies for the discovery of precise medicine against non-small cell lung cancer

Dysregulation of MAPK pathway receptors are crucial in causing uncontrolled cell proliferation in many cancer types including non-small cell lung cancer. Due to the complications in targeting the upstream components, MEK is an appealing target to diminish this pathway activity. Hence, we have aimed to discover potent MEK inhibitors by integrating virtual screening and machine learning-based strategies. Preliminary screening was conducted on 11,808 compounds using the cavity-based pharmacophore model AADDRRR. Further, seven ML models were accessed to predict the MEK active compounds using six molecular representations. The LGB model with morgan2 fingerprints surpasses other models ensuing 0.92 accuracy and 0.83 MCC value versus test set and 0.85 accuracy and 0.70 MCC value with external set. Further, the binding ability of screened hits were examined using glide XP docking and prime-MM/GBSA calculations. Note that we have utilized three ML-based scoring functions to predict the various biological properties of the compounds. The two hit compounds such as DB06920 and DB08010 resulted excellent binding mechanism with acceptable toxicity properties against MEK. Further, 200 ns of MD simulation combined with MM-GBSA/PBSA calculations confirms that DB06920 may have stable binding conformations with MEK thus step forwarded to the experimental studies in the near future. Communicated by Ramaswamy H. Sarma

A machine learning-based strategy for experimentally estimating force chains of granular materials using X-ray micro-tomography

A machine learning-based strategy is presented to estimate the contact force chains of uniformly sized spherical granular materials under triaxial compression, using particle kinematics and inter-particle contact evolution data measured by X-ray micro-tomography (μCT). To this end, a graph neural network (GNN) is introduced to predict the contact force chains of granular materials at the end of a shear increment based on the evolution of contact network and grain displacement during that shear increment. Meanwhile, discrete-element modelling (DEM) is performed for a glass bead specimen under triaxial compression with the use of in situ μCT scanning. The DEM model has the same initial conditions as the glass bead specimen and is validated by comparing the calculated stress–strain curves, particle kinematics and inter-particle contact fabric evolution of the numerical specimen with the experimental results. The DEM model is used to generate sufficient virtual data to train the GNN model, which is applied to the glass bead specimen to predict the evolution of contact force chains. The model-predicted results yield a power-law relationship between the above mean normalised particle maximum normal contact forces and the probability density function, which is consistent with the findings reported in previous numerical studies.

Mapping of disease-specific Oxford Knee Score onto EQ-5D-5L utility index in knee osteoarthritis

BackgroundEQ5D is a generic measure of health. It provides a single index value for health status that can be used in the clinical and economic evaluation of healthcare. Oxford Knee Score (OKS) is a joint-specific outcome measure tool designed to assess symptoms and function in osteoarthritis patients after joint replacement surgery. Though widely used, it has the disadvantage of lacking health index value. To fill the gap between functional and generic questionnaires with economic value, we linked generic EQ-5D-5L to the specific OKS to give a single index value for health status in KOA patients.Questions/purposesDeveloping and evaluating an algorithm to estimate EuroQoL generic health utility scores (EQ-5D-5L) from the disease-specific OKS using data from patients with knee osteoarthritis (KO).Patients and methodsThis is a cross-sectional study of 571 patients with KO. We used four distinct mapping algorithms: Cumulative Probability for Ordinal Data, Penalized Ordinal Regression, CART (Classification and Regression Trees), and Ordinal random forest. We compared the resultant models’ degrees of accuracy.ResultsMobility was best predicted by penalized regression with pre-processed predictors, usual activities by random forest, pain/discomfort by cumulative probability with pre-processed predictors, self-care by random forest with RFE (recursive feature elimination) predictors, and anxiety/depression by CART with RFE predictors. Model accuracy was lowest with anxiety/depression and highest with mobility and usual activities. Using available country value sets, the average MAE was 0.098 ± 0.022, ranging from 0.063 to 0.142; and the average MSE was 0.020 ± 0.008 ranging from 0.008 to 0.042.ConclusionsThe current study derived accurate mapping techniques from OKS to the domains of EQ-5D-5L, allowing for the computation of QALYs in economic evaluations. A machine learning-based strategy offers a viable mapping alternative that merits further exploration.

Machine Learning-Based Control for Fuel Cell Hybrid Buses: From Average Load Power Prediction to Energy Management

In this work, a machine learning-based energy management system is developed using a long short-term memory (LSTM) network for fuel cell hybrid buses. The neural network implicitly learns the complex relationship between various factors and the optimal power control from massive data. The selection of the neural network inputs is inspired by the adaptive Pontryagin’s minimum principle (APMP) strategy. Since an estimated value of the global average fuel cell power is required in the machine learning-based energy management strategy (EMS), some global features of driving cycles are extracted and then applied in a feedforward neural network to predict the average fuel cell power appropriately. The effectiveness of the machine learning-based energy management, with the integration of the mechanism of estimating the average fuel cell power based on the forward neural network, is tested under two different driving cycles from the training environment, with comparisons to a commercially used rule-based strategy. Based on the simulation results, the learning-based strategy outperforms the rule-based strategy regarding the charge-sustaining mode conditions and fuel economy. Moreover, compared to the best offline hydrogen consumption, the machine learning-based strategy consumed 0.58% and 0.36% more than the best offline results for both driving cycles. In contrast, the rule-based strategy consumed 1.80% and 0.96% more than optimal offline results for the two driving cycles, respectively. Finally, simulations under battery and fuel cell aging conditions show that the fuel economy of the machine learning-based strategy experiences no performance degradation under components aging compared to offline strategies.

What drives performance in machine learning models for predicting heart failure outcome?

The development of acute heart failure (AHF) is a critical decision point in the natural history of the disease and carries a dismal prognosis. The lack of appropriate risk-stratification tools at hospital discharge of AHF patients significantly limits clinical ability to precisely tailor patient-specific therapeutic regimen at this pivotal juncture. Machine learning-based strategies may improve risk stratification by incorporating analysis of high-dimensional patient data with multiple covariates and novel prediction methodologies. In the current study, we aimed at evaluating the drivers for success in prediction models and establishing an institute-tailored artificial Intelligence-based prediction model for real-time decision support. We used a cohort of all 10 868 patients AHF patients admitted to a tertiary hospital during a 12 years period. A total of 372 covariates were collected from admission to the end of the hospitalization. We assessed model performance across two axes: (i) type of prediction method and (ii) type and number of covariates. The primary outcome was 1-year survival from hospital discharge. For the model-type axis, we experimented with seven different methods: logistic regression (LR) with either L1 or L2 regularization, random forest (RF), Cox proportional hazards model (Cox), extreme gradient boosting (XGBoost), a deep neural-net (NeuralNet) and an ensemble classifier of all the above methods. We were able to achieve an area under receiver operator curve (AUROC) prediction accuracy of more than 80% with most prediction models including L1/L2-LR (80.4%/80.3%), Cox (80.2%), XGBoost (80.5%), NeuralNet (80.4%). RF was inferior to other methods (78.8%), and the ensemble model was slightly superior (81.2%). The number of covariates was a significant modifier (P < 0.001) of prediction success, the use of multiplex-covariates preformed significantly better (AUROC 80.4% for L1-LR) compared with a set of known clinical covariates (AUROC 77.8%). Demographics followed by lab-tests and administrative data resulted in the largest gain in model performance. The choice of the predictive modelling method is secondary to the multiplicity and type of covariates for predicting AHF prognosis. The application of a structured data pre-processing combined with the use of multiple-covariates results in an accurate, institute-tailored risk prediction in AHF.