Variable Selection Criteria Research Articles

An efficient joint compression and sparsity estimation matching pursuit algorithm for artificial intelligence application

Measurement matrix and signal reconstruction algorithm are the key factors influencing the performance of signal reconstruction. However, the reconstruction performance of the existing matching pursuit algorithms, the most popular reconstruction algorithms, is closely related to the signal sparsity, which is hard to determinate apriori. As well, the researches on the reconstruction algorithms are developed independently with the design of measurement matrix. So, in this paper, we originally conduct a joint study of design of measurement matrix and signal reconstruction algorithm. RIP criterion is used to quantitatively analyze the relationship between the signal sparsity and the measurement matrix, and then an efficient joint compression and sparsity estimation matching pursuit (JCSEMP) algorithm is proposed. JCSEMP algorithm constructs a chaotic measurement matrix, a sparsity estimation algorithm based on the chaotic measurement matrix, and a variable atom selection criterion which use the variation between the residuals to adaptively adjust the number of atoms to select. Experimental results demonstrate that this algorithm can provide a better reconstruction performance and a lower reconstruction period.

A variable selection method for multiclass classification problems using two-class ROC analysis

Modern procedures in analytical chemistry generate enormous amounts of data, which must be processed and interpreted. The treatment of such high-dimensional datasets often necessitates the prior selection of a reduced number of variables in order to extract knowledge about the system under study and to maximize the predictability of the models built. Therefore, this article describes a variable selection method for multiclass classification problems using two-class ROC analysis and its associated area under the ROC curve as a variable selection criterion. The variable selection method has been successfully applied to two datasets. For comparison purposes, two other variable selection methods, ReliefF and mRMR, were used and double cross-validation PLS-DA was applied using: (1) all variables and (2) the variables selected using the three methods. It has been demonstrated that correct variable selection can substantially reduce the dimensionality of the datasets, while maximizing the predictability of the models.

Conditional Akaike information under covariate shift with application to small area estimation

AbstractIn this study, we consider the problem of selecting explanatory variables of fixed effects in linear mixed models under covariate shift, which is when the values of covariates in the model for prediction differ from those in the model for observed data. We construct a variable selection criterion based on the conditional Akaike information introduced by Vaida & Blanchard (2005). We focus especially on covariate shift in small area estimation and demonstrate the usefulness of the proposed criterion. In addition, numerical performance is investigated through simulations, one of which is a design‐based simulation using a real dataset of land prices. The Canadian Journal of Statistics 46: 316–335; 2018 © 2018 Statistical Society of Canada

Consistent and robust variable selection in regression based on Wald test

ABSTRACTSelection of relevant predictor variables for building a model is an important problem in the multiple linear regression. Variable selection method based on ordinary least squares estimator fails to select the set of relevant variables for building a model in the presence of outliers and leverage points. In this article, we propose a new robust variable selection criterion for selection of relevant variables in the model and establish its consistency property. Performance of the proposed method is evaluated through simulation study and real data.

Prevalence of household falls in long-lived adults and association with extrinsic factors.

ResumeObjective:to identify the prevalence of falls among older adults and the extrinsic factors associated with them. Method:population-based cross-sectional study with 350 older adults. A household survey was conducted using a questionnaire addressing socio-demographic, clinical, and environmental characteristics. Data were analyzed using Stata Software V.10. Pearson’s chi-square test and logistic regression analysis were used with stepwise criteria for selection of variables in the model, with measures of effect expressed in Prevalence Ratio. For input into the multiple model, the variables with p ≤ 0.20 were considered. All ethical care regarding research on human beings has been observed and respected. Results:the prevalence of falls was 46.9%. The extrinsic factors associated with falls were: stairs, uneven floor and pets in the main entrance, lack of anti-slip loose throw rugs and slippery floor in the kitchen, lack of anti-slip loose throw rugs and objects on the floor in the room, lack of grab bars in the shower, lack of grab bars in the toilet and switch away from the bathroom door (p <0.05). Conclusion:falls are frequent in long-lived adults. The identification of the extrinsic factors associated with the occurrence of this event can help in its prevention.

Selection of robust variables for transfer of classification models employing the successive projections algorithm

Multivariate models have been widely used in analytical problems involving quantitative and qualitative analyzes. However, there are cases in which a model is not applicable to spectra of samples obtained under new experimental conditions or in an instrument not involved in the modeling step. A solution to this problem is the transfer of multivariate models, usually performed using standardization of the spectral responses or enhancement of the robustness of the model. This present paper proposes two new criteria for selection of robust variables for classification transfer employing the successive projections algorithm (SPA). These variables are then used to build models based on linear discriminant analysis (LDA) with low sensitivity with respect to the differences between the responses of the instruments involved. For this purpose, transfer samples are included in the calculation of the cost for each subset of variables under consideration. The proposed methods are evaluated for two case studies involving identification of adulteration of extra virgin olive oil (EVOO) and hydrated ethyl alcohol fuel (HEAF) using UV–Vis and NIR spectroscopy, respectively. In both cases, similar or better classification transfer results (obtained for a test set measured on the secondary instrument) employing the two criteria were obtained in comparison with direct standardization (DS) and piecewise direct standardization (PDS). For the UV–Vis data, both proposed criteria achieved the correct classification rate (CCR) of 85%, while the best CCR obtained for the standardization methods was 81% for DS. For the NIR data, 92.5% of CCR was obtained by both criteria as well as DS. The results demonstrated the possibility of using either of the criteria proposed for building robust models as an alternative to the standardization of spectral responses for transfer of classification.

Neural network models for supporting drug and multidrug resistant tuberculosis screening diagnosis

Tuberculosis (TB) is the leading cause of global mortality among communicable diseases. The diagnosis of Drug-Resistant Tuberculosis (DR-TB) demands even more attention, leading to longer treatments and higher deceased rates. All diagnostic methods available have deficiencies in their detection rates, time release results, or have a higher cost and need a complex infrastructure to setup. New molecular diagnostics, such as the Xpert MTB/RIF assay, have great potential for revolutionizing the diagnosis of Rifampicin Resistance (RR). But, a positive RR result with this test should be carefully interpreted and take into consideration the risk of Multidrug-Resistant TB (MDR-TB) according to its prevalence, locally. Therefore, the development of screening approaches for DR/MDR-TB suspects would help to identify those should be tested by Xpert MTB/RIF. This work develops Artificial Neural Network (ANN) models considering data from presumed DR/MDR-TB subjects according to the National Guidelines at Rio de Janeiro/Brazil, attended in reference centers in Rio de Janeiro, from Feb 2011 and May 2013. Subjects aged 18 years or older, and results were compared with models based on Classification And Regression Trees (CART). Practical operation at different epidemiological scenarios are considered by constructing models using different variable selection criteria, so that environments with low resource conditions can be assisted. Among 280 presumed DR-TB cases included, 38 were DR-TB, 48-MDR, 32-Drug-Sensitive and 162 with no TB. Between DR-TB and non DR-TB, the sensitivity and specificity reached 95.1%(±5.0) and 85.0%(±4.9), respectively. The promising results of clinical score for DR/MDR-TB diagnosis indicate that this approach may be used in the evaluation of presumed DR/MDR-TB.

Do consensus models outperform individual models? Transferability evaluations of diverse modeling approaches for an invasive moth

Transferability is key to many of the most novel and interesting applications of ecological niche models, such that maximizing predictive power of model transfers is crucial. Here, we explored consensus methods as a means of reducing uncertainty and improving model transferability in anticipating the potential distribution of an invasive moth (Hyphantria cunea). Individual native-range niche models were calibrated using seven modelling algorithms and four environmental datasets, representing different degrees of dimensionality, spatial correlation, and ecological relevance, and showing different degrees of climate niche expansion. Four consensus methods were used to combine individual niche models; we assessed transferability of consensus models and the individual models used to generate them. The results suggested that ideal criteria for environmental variable selection vary among algorithms, as different algorithms showed different sensitivities to spatial dimensionality and correlation. Consensus models reflected the central tendency of individual models, and reduced uncertainty by consolidating consistency across individual models, but did not outperform individual models. The question of whether interpolation accuracy comes at the expense of transferability suggests caution in planning methodologies for processing niche models to predict invasive potential. These explorations outline approaches by which to reduce uncertainty and improve niche model transferability with vital implications for ensemble forecasting.

Focused Information Criterion for Series Estimation in Partially Linear Models

This paper proposes a focused information criterion for variable selection in partially linear models. Our criterion is designed to select an optimal model for estimating a focus parameter, which is a parameter of interest. We estimate the model using the series method and jointly select the variables in the linear part and the series length in the nonparametric part. A Monte Carlo simulation shows that the proposed focused information criterion successfully selects the model that has a relatively small mean squared error of the estimator for the focus parameter.

Simultaneous Confidence Corridors and Variable Selection for Generalized Additive Models

In spite of the widespread use of generalized additive models (GAMs), there is no well established methodology for simultaneous inference and variable selection for the components of GAM. There is no doubt that both, inference on the marginal component functions and their selection, are essential in this additive statistical models. To this end, we establish simultaneous confidence corridors (SCCs) and a variable selection criteria through the spline-backfitted kernel smoothing techniques. To characterize the global features of each component, SCCs are constructed for testing their shapes. By extending the BIC to additive models with identity/trivial link, an asymptotically consistent BIC approach for variable selection is proposed. Our procedures are examined in simulations for its theoretical accuracy and performance, and used to forecast the default probability of listed Japanese companies.

Big Data, Predictive Analytics, and Quality Improvement in Kidney Transplantation: A Proof of Concept.

We sought proof of concept of a Big Data Solution incorporating longitudinal structured and unstructured patient-level data from electronic health records (EHR) to predict graft loss (GL) and mortality. For a quality improvement initiative, GL and mortality prediction models were constructed using baseline and follow-up data (0-90 days posttransplant; structured and unstructured for 1-year models; data up to 1 year for 3-year models) on adult solitary kidney transplant recipients transplanted during 2007-2015 as follows: Model 1: United Network for Organ Sharing (UNOS) data; Model 2: UNOS & Transplant Database (Tx Database) data; Model 3: UNOS, Tx Database & EHR comorbidity data; and Model 4: UNOS, Tx Database, EHR data, Posttransplant trajectory data, and unstructured data. A 10% 3-year GL rate was observed among 891 patients (2007-2015). Layering of data sources improved model performance; Model 1: area under the curve (AUC), 0.66; (95% confidence interval [CI]: 0.60, 0.72); Model 2: AUC, 0.68; (95% CI: 0.61-0.74); Model 3: AUC, 0.72; (95% CI: 0.66-077); Model 4: AUC, 0.84, (95 % CI: 0.79-0.89). One-year GL (AUC, 0.87; Model 4) and 3-year mortality (AUC, 0.84; Model 4) models performed similarly. A Big Data approach significantly adds efficacy to GL and mortality prediction models and is EHR deployable to optimize outcomes.

Direct oral anticoagulants for the treatment of cancer-associated venous thromboembolism.

Cancer patients with venous thromboembolism (VTE) are at increased risk for both bleeding and VTE recurrence. Anticoagulation with low-molecular-weight heparin (LMWH) is the standard of care during the initial and longterm treatment phase (i.e. during the first 3 - 6 months of therapy) based on its overall beneficial safety and efficacy profile compared to vitamin K antagonists (VKAs). The direct oral anticoagulants (DOACs) rivaroxaban, apixaban, edoxaban, and dabigatran are approved for the treatment of acute VTE, and the combined six phase-3 trials have included > 1500 patients with active cancer, as defined by variable selection criteria. Subgroup analyses of these patients, either pooled or separately reported, suggest that DOACs could be a safe and efficacious alternative to VKA therapy for the treatment of cancer-associated VTE. However, the populations of cancer patients included in the DOAC and LMWH trials are not comparable with regard to mortality and VTE risk, and no specific data from direct head-to-head comparisons of DOACs with LMWHs are currently available. The use of DOACs for the management of VTE in cancer is thus not recommended by clinical practice guidelines.

Direct oral anticoagulants for the treatment of cancer-associated venous thromboembolism

SummaryCancer patients with venous thromboembolism (VTE) are at increased risk for both bleeding and VTE recurrence. Anticoagulation with low-molecular-weight heparin (LMWH) is the standard of care during the initial and longterm treatment phase (i.e. during the first 3–6 months of therapy) based on its overall beneficial safety and efficacy profile compared to vitamin K antagonists (VKAs). The direct oral anticoagulants (DOACs) rivaroxaban, apixaban, edoxaban, and dabigatran are approved for the treatment of acute VTE, and the combined six phase-3 trials have included &gt; 1 500 patients with active cancer, as defined by variable selection criteria. Subgroup analyses of these patients, either pooled or separately reported, suggest that DOACs could be a safe and efficacious alternative to VKA therapy for the treatment of cancer-associated VTE. However, the populations of cancer patients included in the DOAC and LMWH trials are not comparable with regard to mortality and VTE risk, and no specific data from direct head-to-head comparisons of DOACs with LMWHs are currently available. The use of DOACs for the management of VTE in cancer is thus not recommended by clinical practice guidelines.

A robust and efficient stepwise regression method for building sparse polynomial chaos expansions

Polynomial Chaos (PC) expansions are widely used in various engineering fields for quantifying uncertainties arising from uncertain parameters. The computational cost of classical PC solution schemes is unaffordable as the number of deterministic simulations to be calculated grows dramatically with the number of stochastic dimension. This considerably restricts the practical use of PC at the industrial level. A common approach to address such problems is to make use of sparse PC expansions. This paper presents a non-intrusive regression-based method for building sparse PC expansions. The most important PC contributions are detected sequentially through an automatic search procedure. The variable selection criterion is based on efficient tools relevant to probabilistic method. Two benchmark analytical functions are used to validate the proposed algorithm. The computational efficiency of the method is then illustrated by a more realistic CFD application, consisting of the non-deterministic flow around a transonic airfoil subject to geometrical uncertainties. To assess the performance of the developed methodology, a detailed comparison is made with the well established LAR-based selection technique. The results show that the developed sparse regression technique is able to identify the most significant PC contributions describing the problem. Moreover, the most important stochastic features are captured at a reduced computational cost compared to the LAR method. The results also demonstrate the superior robustness of the method by repeating the analyses using random experimental designs.

Stochastic correlation coefficient ensembles for variable selection

ABSTRACTIn this paper, we propose a novel Max-Relevance and Min-Common-Redundancy criterion for variable selection in linear models. Considering that the ensemble approach for variable selection has been proven to be quite effective in linear regression models, we construct a variable selection ensemble (VSE) by combining the presented stochastic correlation coefficient algorithm with a stochastic stepwise algorithm. We conduct extensive experimental comparison of our algorithm and other methods using two simulation studies and four real-life data sets. The results confirm that the proposed VSE leads to promising improvement on variable selection and regression accuracy.

Bayesian Ensemble Trees (BET) for Clustering and Prediction in Heterogeneous Data

We propose a novel “tree-averaging” model that uses the ensemble of classification and regression trees (CART). Each constituent tree is estimated with a subset of similar data. We treat this grouping of subsets as Bayesian ensemble trees (BET) and model them as a Dirichlet process. We show that BET determines the optimal number of trees by adapting to the data heterogeneity. Compared with the other ensemble methods, BET requires much fewer trees and shows equivalent prediction accuracy using weighted averaging. Moreover, each tree in BET provides variable selection criterion and interpretation for each subset. We developed an efficient estimating procedure with improved estimation strategies in both CART and mixture models. We demonstrate these advantages of BET with simulations and illustrate the approach with a real-world data example involving regression of lung function measurements obtained from patients with cystic fibrosis. Supplementary materials for this article are available online.

Minimizing variable selection criteria by Markov chain Monte Carlo

Regression models with a large number of predictors arise in diverse fields of social sciences and natural sciences. For proper interpretation, we often would like to identify a smaller subset of the variables that shows the strongest information. In such a large size of candidate predictors setting, one would encounter a computationally cumbersome search in practice by optimizing some criteria for selecting variables, such as AIC, $$C_{P}$$CP and BIC, through all possible subsets. In this paper, we present two efficient optimization algorithms vis Markov chain Monte Carlo (MCMC) approach for searching the global optimal subset. Simulated examples as well as one real data set exhibit that our proposed MCMC algorithms did find better solutions than other popular search methods in terms of minimizing a given criterion.

Morphological characterization of breeds of sheep: a discriminant analysis approach.

In this study the characterisation and separation/discrimination of three sheep breeds (crosses, West African Dwarfs (WAD) and West African Long Legged (WALL)] based on their physical traits (morphological characterisation) was investigated extensively with the application of discriminant analysis. The study’s main objective was specifically based on developing a variable selection criterion that can discriminate best among the three sheep breeds and as well as obtain a reliable mathematical function/equation (discriminant functions) for provision of maximum separation among the three known sheep breeds. Data from College of Education, Mampong animal farms on various breeds of sheep (hybrid/crossed breed, Sahell/WALL and Djallonke/WAD) was used. Factor Analysis was employed as a variable selection criterion for selecting six sheep traits that can discriminate best among the sheep breeds. Canonical discriminant function was derived for the eight variable data set and was compared with the derived quadratic discriminant functions (QDFs) using the six extracted sheep traits. The six variable QDF distance classifier provided maximum separation after cross validation than the 8-variable canonical discriminant functions. The derived mathematical functions (QDFs) were able to provide maximum separation among the three known sheep breeds with a correct classification rate of 0.86.

Do the Adult Daughters of PCOS Patients Develop PCOS and Is This Due to an Androgenized Uterine Environment-An Online Epidemiological Survey

Objectives: Several inconsistent studies have investigated whether the uterine environment of androgenized pregnant women is a risk factor for an in-utero developmental imprinted predisposition towards subsequent polycystic ovarian syndrome (PCOS) among their female offspring. These are difficult to compare due to variable parameters and subject selection criteria. Few epidemiological studies have analyzed the incidence of PCOS amongst adult daughters of PCOS affected women previously. Our study aimed to investigate risk factors relating to the development of PCOS in the female offspring of PCOS patients. Methods: We used a questionnaire to collect a mother-to-daughter medical history and relevant information, in order to understand risk factors, which might relate to the presence of PCOS daughters of PCOS patients. Results: Of four hundred and one responses, 131 participants were included in the final analysis. There was no statistical association with the subsequent development of PCOS amongst female offspring of women with PCOS. However, there was a significantly higher prevalence of post-term birth among PCOS mothers. Nevertheless, the major determinant of risk of subsequent incidence of PCOS amongst daughters was a higher BMI, regardless of the mothers BMI. Conclusion: Socio-economic family influences, affecting BMI, may be the reason for any mother to daughter association with PCOS.

Model selection in the weighted generalized estimating equations for longitudinal data with dropout.

We propose criteria for variable selection in the mean model and for the selection of a working correlation structure in longitudinal data with dropout missingness using weighted generalized estimating equations. The proposed criteria are based on a weighted quasi-likelihood function and a penalty term. Our simulation results show that the proposed criteria frequently select the correct model in candidate mean models. The proposed criteria also have good performance in selecting the working correlation structure for binary and normal outcomes. We illustrate our approaches using two empirical examples. In the first example, we use data from a randomized double-blind study to test the cancer-preventing effects of beta carotene. In the second example, we use longitudinal CD4 count data from a randomized double-blind study.