关键词： artificial neural network   fertility traits   genomic selection   random forest   support vector regression  

摘要： This study aimed to assess the predictive ability of different machine learning (ML) methods for genomic prediction of reproductive traits in Nellore cattle. The studied traits were age at first calving (AFC), scrotal circumference (SC), early pregnancy (EP) and stayability (STAY). The numbers of genotyped animals and SNP markers available were 2342 and 321 419 (AFC), 4671 and 309 486 (SC), 2681 and 319 619 (STAY) and 3356 and 319 108 (EP). Predictive ability of support vector regression (SVR), Bayesian regularized artificial neural network (BRANN) and random forest (RF) were compared with results obtained using parametric models (genomic best linear unbiased predictor, GBLUP, and Bayesian least absolute shrinkage and selection operator, BLASSO). A 5-fold cross-validation strategy was performed and the average prediction accuracy (ACC) and mean squared errors (MSE) were computed. The ACC was defined as the linear correlation between predicted and observed breeding values for categorical traits (EP and STAY) and as the correlation between predicted and observed adjusted phenotypes divided by the square root of the estimated heritability for continuous traits (AFC and SC). The average ACC varied from low to moderate depending on the trait and model under consideration, ranging between 0.56 and 0.63 (AFC), 0.27 and 0.36 (SC), 0.57 and 0.67 (EP), and 0.52 and 0.62 (STAY). SVR provided slightly better accuracies than the parametric models for all traits, increasing the prediction accuracy for AFC to around 6.3 and 4.8% compared with GBLUP and BLASSO respectively. Likewise, there was an increase of 8.3% for SC, 4.5% for EP and 4.8% for STAY, comparing SVR with both GBLUP and BLASSO. In contrast, the RF and BRANN did not present competitive predictive ability compared with the parametric models. The results indicate that SVR is a suitable method for genome-enabled prediction of reproductive traits in Nellore cattle. Further, the optimal kernel bandwidth parameter in the SV

关键词： Medical imaging   Retinal vessels   Segmentation   Machine learning   Deep learning   Review  

摘要： The eye affords a unique opportunity to inspect a rich part of the human microvasculature non-invasively via retinal imaging. Retinal blood vessel segmentation and classification are prime steps for the diagnosis and risk assessment of microvascular and systemic diseases. A high volume of techniques based on deep learning have been published in recent years. In this context, we review 158 papers published between 2012 and 2020, focussing on methods based on machine and deep learning (DL) for automatic vessel segmentation and classification for fundus camera images. We divide the methods into various classes by task (segmentation or artery-vein classification), technique (supervised or unsupervised, deep and non-deep learning, hand-crafted methods) and more specific algorithms (e.g. multiscale, morphology). We discuss advantages and limitations, and include tables summarising results at-a-glance. Finally, we attempt to assess the quantitative merit of DL methods in terms of accuracy improvement compared to other methods. The results allow us to offer our views on the outlook for vessel segmentation and classification for fundus camera images. (c) 2020 Elsevier B.V. All rights reserved.

关键词： groundwater storage   terrestrial water storage   downscaling   random forest   XGBoost   GRACE   GLDAS  

摘要： High-resolution and continuous hydrological products have tremendous importance for the prediction of water-related trends and enhancing the capability for sustainable water resources management under climate change and human impacts. In this study, we used the random forest (RF) and extreme gradient boosting (XGBoost) methods to downscale groundwater storage (GWS) from 1 degrees (similar to 110 km) to 1 km by downscaling Gravity Recovery and Climate Experiment (GRACE) and Global Land Data Assimilation System (GLDAS) data from 1 degrees (similar to 110 km) and 0.25 degrees (similar to 25 km) respectively, to 1 km for China. Three evaluation metrics were employed for the testing dataset for 2004-2016: The R-2 ranged from 0.77-0.89 for XGBoost (0.74-0.86 for RF), the correlation coefficient (CC) ranged from 0.88-0.94 for XGBoost (0.88-0.93 for RF) and the root-mean-square error (RMSE) ranged from 0.37-2.3 for XGBoost (0.4-2.53 for RF). The R-2 of the XGBoost models for GLDAS was 0.64-0.82 (0.63-0.82 for RF), the CC was 0.80-0.91 (0.80-0.90 for RF) and the RMSE was 0.63-1.75 (0.63-1.77 for RF). The downscaled GWS derived from GRACE and GLDAS were validated using in situ measurements by comparing the time series variations and the downscaled products maintained the accuracy of the original data. The interannual changes within 9 river basins between pre- and post-downscaling were consistent, emphasizing the reliability of the downscaled products. Ultimately, annual downscaled TWS, GLDAS and GWS products were provided from 2004 to 2016, providing a solid data foundation for studying local GWS changes, conducting finer-scale hydrological studies and adapting water resources management and policy formulation to local condition.

关键词： FEM   Regression   Buckling   Composite laminate   MARS   MPMR   Skew  

摘要： Purpose The purpose of this paper is to attempt the buckling analysis of a laminated composite skew plate using the C(0)finite element (FE) model based on higher-order shear deformation theory (HSDT) in conjunction with minimax probability machine regression (MPMR) and multivariate adaptive regression spline (MARS). Design/methodology/approach HSDT considers the third-order variation of in-plane displacements which eliminates the use of shear correction factor owing to realistic parabolic transverse shear stresses across the thickness coordinate. At the top and bottom of the plate, zero transverse shear stress condition is imposed. C0FE model based on HSDT is developed and coded in formula translation (FORTRAN). FE model is validated and found efficient to create new results. MPMR and MARS models are coded in MATLAB. Using skew angle (alpha), stacking sequence (Ai) and buckling strength (Y) as input parameters, a regression problem is formulated using MPMR and MARS to predict the buckling strength of laminated composite skew plates. Findings The results of the MPMR and MARS models are in good agreement with the FE model result. MPMR is a better tool than MARS to analyze the buckling problem. Research limitations/implications The present work considers the linear behavior of the laminated composite skew plate. Originality/value To the authors' best of knowledge, there is no work in the literature on the buckling analysis of a laminated composite skew plate using C0FE formulation based on third-order shear deformation theory in conjunction with MPMR and MARS. These machine-learning techniques increase efficiency, reduce the computational time and reduce the cost of analysis. Further, an equation is generated with the MARS model via which the buckling strength of the laminated composite skew plate can be predicted with ease and simplicity.

关键词： Fetal birthweight prediction   Health data mining   Pregnant healthcare   Temporal data mining  

摘要： Background Birthweight is an important indicator during the fetal development process to protect the maternal and infant safety. However, birthweight is difficult to be directly measured, and is usually roughly estimated by the empirical formulas according to the experience of the doctors in clinical practice. Methods This study attempts to combine multiple electronic medical records with the B-ultrasonic examination of pregnant women to construct a hybrid birth weight predicting classifier based on long short-term memory (LSTM) networks. The clinical data were collected from 5,759 Chinese pregnant women who have given birth, with more than 57,000 obstetric electronic medical records. We evaluated the prediction by the mean relative error (MRE) and the accuracy rate of different machine learning classifiers at different predicting periods for first delivery and multiple deliveries. Additionally, we evaluated the classification accuracies of different classifiers respectively for the Small-for-Gestational-age (SGA), Large-for-Gestational-Age (LGA) and Appropriate-for-Gestational-Age (AGA) groups. Results The results show that the accuracy rate of the prediction model using Convolutional Neuron Networks (CNN), Random Forest (RF), Linear-Regression, Support Vector Regression (SVR), Back Propagation Neural Network(BPNN), and the proposed hybrid-LSTM at the 40th pregnancy week for first delivery were 0.498, 0.662, 0.670, 0.680, 0.705 and 0.793, respectively. Among the groups of less than 39th pregnancy week, the 39th pregnancy week and more than 40th week, the hybrid-LSTM model obtained the best accuracy and almost the least MRE compared with those of machine learning models. Not surprisingly, all the machine learning models performed better than the empirical formula. In the SGA, LGA and AGA group experiments, the average accuracy by the empirical formula, logistic regression (LR), BPNN, CNN, RF and Hybrid-LSTM were 0.780, 0.855, 0.890, 0.906, 0.916 and 0.933, respectiv

关键词： Terrain identification   Foot-to-ground angle   Support vector machine   Convolution neural network  

摘要： This paper deals with the identification of terrain that is crucial to trigger the damping in semi-active lower limb prosthesis. Objective: To identify level ground and ramp terrains using foot-to-ground angle (FGA) measurement. Methods: First, Instrumented shoe for FGA measurement was developed. Next, data collection from able-bodied (n = 5) and amputee (n = 1) participants was carried out. Finally, a comparison of identification accuracy using support vector machine (SVM) and convolution neural network (CNN) algorithms was done. Results: The average classification accuracy obtained for able-bodied participants and amputee is 79.57% +/- 20.32% and 73.06% +/- 12.70%, respectively using SVM, whereas it is 83.45% +/- 14.50% and 80% +/- 12.15% respectively using CNN. Our off-line analysis shows that overall, CNN outperformed SVM with an average of 4.86% increment in classification accuracy in able-bodied participants and 9.54% in an amputee. This study introduced a simplified, low-cost method for terrain identification in the prosthetic control application.

关键词： Proximal hyperspectral imaging   Tobacco plant   Heavy metal Hg stress   Canopy   Machine learning methods  

摘要： Accurate detection of heavy metal stress on the growth status of plants is of great concern for agricultural production and management, food security, and ecological environment. A proximal hyperspectral imaging (HSI) system covered the visible/near-infrared (Vis/NIR) region of 400-1000 nm coupled with machine learning methods were employed to discriminate the tobacco plants stressed by different concentration of heavy metal Hg. After acquiring hyperspectral images of tobacco plants stressed by heavy metal Hg with concentration solutions of 0 mg.L-1 (non-stressed groups), 1, 3, and 5 mg.L-1 (3 stressed groups), regions of interest (ROIs) of canopy in tobacco plants were identified for spectra processing. Meanwhile, tobacco plant's appearance and microstructure of mesophyll tissue in tobacco leaves were analyzed. After that, clustering effects of the non-stressed and stressed groups were revealed by score plots and score images calculated by principal component analysis (PCA). Then, loadings of PCA and competitive adaptive reweighted sampling (CARS) algorithm were employed to pick effective wavelengths (EWs) for discriminating non-stressed and stressed samples. Partial least squares discriminant analysis (PLS-DA) and least-squares support vector machine (LS-SVM) were utilized to estimate the stressed tobacco plants status with different concentrations Hg solutions. The performances of those models were evaluated using confusion matrixes (CMes) and receiver operating characteristics (ROC) curves. Results demonstrated that PLS-DA models failed to offer relatively good result, and this algorithm was abandoned to classify the stressed and non-stressed groups of tobacco plants. Compared to LS-SVM model based on full spectra (FS-LS-SVM), the LS-SVM model established EWs selected by CARS (CARS-LS-SVM) carried 13 variables provided an accuracy of 100%, which was promising to achieve the qualitative discrimination of the non-stressed and stressed tobacco plants. Meanwhile, fo

关键词： Android Malware Family   Multi-classification   Machine learning   Transfer Learning  

摘要： In order to overcome the lasting increase of Android malware, malware family classification, which clusters malware with the same features into one family, has been proposed as an efficient way for malware analysis. Several machine learning based approaches have been proposed for such task of malware family classification. However, due to the adoption of very different features and learning methods in different approaches, it is still an open question to explore: which approach works better for malware family classification? In this paper, we conduct extensive experiments to answer this question. For three widely known Android malware datasets, we design five multi-classification methods for predicting Android malware family. Based on the survey of Android malware analysis literatures and the observation of a large number of Android malware, we construct a set of 250 common features shared by Android malware. And we also collect 16873 documentary features from Android Developer as a comparison. Furthermore, we investigate the effects of transfer learning for adapting the model on three malware datasets on different scales. Our empirical results show that (i) the classification methods perform very closely, with neural network model having marginally better performance (1% to 3% in F1-score), (ii) features contribute most for classification, especially to enhance API features on larger datasets, and (iii) it is model transferable across different malware datasets based on various transfer learning tasks.

关键词： Wastewater   Machine learning   Dissolved organic matter   Basicity   Water treatment  

摘要： Oxidation of micropollutants (MPs) by ozonation proceeds via the reactions with molecular ozone (O-3) and hydroxyl radicals ((OH)-O-center dot). To predict MP abatement during ozonation, a model that can accurately predict oxidant exposures (i.e., integral(t)(0)[O-3]dt and integral(t)(0)[(OH)-O-center dot]dt) needs to be developed. This study demonstrates machine learning models based on the random forest (RF) algorithm to output oxidant exposures from water quality parameters (input variables) that include pH, alkalinity, dissolved organic carbon concentration, and fluorescence excitation-emission matrix (FEEM) data (to characterize organic matter). To develop the models, 60 different samples of natural waters and wastewater effluents were collected and characterized, and the oxidant exposures in each sample were determined at a specific O-3 dose (2.5 mg/L). Four RF models were developed depending on how FEEM data were utilized (i.e., one model free of FEEM data, and three other models that used FEEM data of different resolutions). The regression performance and Akaike information criterion (MC) were evaluated for each model. The models using high-resolution FEEM data generally exhibited high prediction accuracy with reasonable AIC values, implying that organic matter characteristics quantified by FEEM can be important factors to improve the accuracy of the prediction model. The developed models can be applied to predict the abatement of MPs in drinking water and wastewater ozonation processes and to optimize the O-3 dose for the intended removal of target MPs. The machine learning models using higher-resolution FEEM data offer more accurate prediction by better calculating the complex nonlinear relationship between organic characteristics and oxidant exposures.

关键词： Powder compaction   Young's modulus   Machine learning   Non-destructive   Material profiling  

摘要： The ability to predict mechanical properties of compacted powder blends of Active Pharmaceutical Ingredients (API) and excipients solely from component properties can reduce the amount of 'trial-and-error' involved in formulation design. Machine Learning (ML) can reduce model development time and effort with the imperative of adequate historical data. This work describes the utility of linear and nonlinear ML models for predicting Young's modulus (YM) of directly compressed blends of known excipients and unknown API mixed at arbitrary compositions given only the true density of the API. The models were trained with data from compacts of three BCS Class I APIs and two excipients blended at four drug loadings, three excipient compositions, and compacted to five nominal solid fractions. The prediction accuracy of the models was measured using three cross-validation (CV) schemes. Finally, we demonstrate an application of the model to enable Quality-by-Design in formulation design. Limitations of the models and future work have also been discussed.