Variable Selection Process Research Articles

Predicting the concentration of indoor culturable fungi using a kernel-based extreme learning machine (K-ELM)

ABSTRACTIndoor fungal is of great significance for human health. The kernel-based extreme learning machine is employed to determine the most important parameters for predicting the concentration of indoor culturable fungi (ICF). For model training and statistical analysis, parameters that contained indoor or outdoor PM10 and PM2.5, RH, Temperature, CO2 and ICF were measured in 85 residential buildings of Baoding, China, from November 2016 to March 2017. The variable selection process contains four different cases to identify the optimal input combination. The results indicate that root mean square error of the optimal input combinations can be improved 5.6% from 1 to 2 input variables, while that could be only improved 1.9% from 2 to 3 input variables. However, considering both precision and simplicity, the combination of indoor PM10 and RH provides a more suitable selection for predicting the ICF.

Clinical factors related to morbidity and mortality in high-risk heart failure patients: the GUIDE-IT predictive model and risk score.

Most heart failure (HF) risk scores have been derived from cohorts of stable HF patients and may not incorporate up to date treatment regimens or deep phenotype characterization that change baseline risk over the short- and long-term follow-up period. We undertook the current analysis of participants in the GUIDE-IT (Guiding Evidence-Based Therapy Using Biomarker Intensified Treatment) trial to address these limitations. The GUIDE-IT study randomized 894 high-risk patients with HF and reduced ejection fraction (≤ 40%) to biomarker-guided treatment strategy vs. usual care. We performed risk modelling using Cox proportional hazards models and analysed the relationship between 35 baseline clinical factors and the primary composite endpoint of cardiovascular (CV) death or HF hospitalization, the secondary endpoint of all-cause mortality, and the exploratory endpoint of 90-day HF hospitalization or death. Prognostic relationships for continuous variables were examined and key predictors were identified using a backward variable selection process. Predictive models and risk scores were developed. Over a median follow-up of 15 months, the cumulative number of HF hospitalizations and CV deaths was 328 out of 894 patients (Kaplan-Meier event rate 34.5% at 12 months). Frequency of all-cause deaths was 143 out of 894 patients (Kaplan-Meier event rate 12.2% at 12 months). Outcomes for the primary and secondary endpoints between strategy arms of the study were similar. The most important predictor that was present in all three models was the baseline natriuretic peptide level. Hispanic ethnicity, low sodium and high heart rate were present in two of the three models. Other important predictors included the presence or absence of a device, New York Heart Association class, HF duration, black race, co-morbidities (sleep apnoea, elevated creatinine, ischaemic heart disease), low blood pressure, and a high congestion score. Risk models using readily available clinical information are able to accurately predict short- and long-term CV events and may be useful in optimizing care and enriching patients for clinical trials. ClinicalTrials.gov ID number NCT01685840.

On integer and MPCC representability of affine sparsity

In addition to sparsity, practitioners of statistics and machine learning often wish to promote additional structures in their variable selection process to incorporate prior knowledge. Borrowing the modeling power of linear systems with binary variables, many of such structures can be faithfully modeled by affine sparsity constraints (ASC). In this note we study conditions under which an ASC system can be represented by sets in integer programs and mathematical programs with complementarity conditions (MPCC). Results of this note facilitate developing nonconvex optimization methods for variable selection with structured sparsity.

A Bayesian Model Averaging approach for modelling tree mortality in relation to site, competition and climatic factors for Chinese fir plantations

Relationships between tree mortality and endogenous factors and climate factors have emerged as important concerns, and logistic stepwise regression is widely used for modeling the relationships. However, this method subsequently ignores both the variables not selected because of insignificance, and the model uncertainty due to the variable selection process. Bayesian Model Averaging (BMA) selects all possible models and uses the posterior probabilities of these models to perform all inferences and predictions. In this study, Bayesian Model Averaging (BMA) and logistic stepwise regression were used to analyze tree mortality in relation to competition, site index, and climatic factors in Chinese fir (Cunninghamia lanceolata (Lamb.) plantations established at five initial planting densities (A: 1667, B: 3333, C: 5000, D: 6667, and E: 10,000 trees/ha). Results showed that the posterior probability of the best model acquired by stepwise regression was less than that of the best model (highest posterior probability) acquired by BMA for pooling the data and density level D. Especially in the other planting densities, the model selected by stepwise regression was not in the BMA models. It indicates that the BMA method performed better than logistic stepwise regression, because BMA gave accurate posterior probability by taking into account the uncertainty of the model. In addition, the mortality increased with high competition and decreased with increasing temperature. The research has important implications for managing Chinese fir plantations under climate change.

Exploration of the variability of variable selection based on distances between bootstrap sample results

It is well known that variable selection in multiple regression can be unstable and that the model uncertainty can be considerable. The model uncertainty can be quantified and explored by bootstrap resampling, see Sauerbrei et al. (Biom J 57:531–555, 2015). Here approaches are introduced that use the results of bootstrap replications of the variable selection process to obtain more detailed information about the data. Analyses will be based on dissimilarities between the results of the analyses of different bootstrap samples. Dissimilarities are computed between the vector of predictions, and between the sets of selected variables. The dissimilarities are used to map the models by multidimensional scaling, to cluster them, and to construct heatplots. Clusters can point to different interpretations of the data that could arise from different selections of variables supported by different bootstrap samples. A new measure of variable selection instability is also defined. The methodology can be applied to various regression models, estimators, and variable selection methods. It will be illustrated by three real data examples, using linear regression and a Cox proportional hazards model, and model selection by AIC and BIC.

Development of a Consecutive Occupancy Estimation Framework for Improving the Energy Demand Prediction Performance of Building Energy Modeling Tools

To improve the energy prediction performance of a building energy model, the occupancy status information is very important. This is more important in real buildings, rather than under construction buildings, because actual building occupancy can significantly influence its energy consumption. In this study, a machine learning based framework for a consecutive occupancy estimation is proposed by utilizing internet of things data, such as indoor temperature and luminance, CO2 density, electricity consumption of lighting, HVAC (heating, ventilation, and air conditioning), electric appliances, etc. Three machine learning based occupancy estimation algorithms (decision tree, support vector machine, artificial neural networks) are selected and evaluated in terms of the performance of estimating the occupancy status for each season. The selection process of the input variables that have crucial impact on the algorithms’ performance are described in detail. Finally, an occupancy estimation framework that can repeat model training and estimation consecutively in a situation when time-series data are continuously provided over the entire measurement period is suggested. In addition, the performance of the framework is evaluated to identify how it improves the energy prediction performance of the building energy model compared to conventional energy modeling practices. The suggested framework is distinguished from similar previous studies in two ways: (1) The proposed framework reveals that input variables for the occupancy estimation model can be occasionally changed by an occupant response to certain times and seasons, and (2) the framework incorporates time-series indirect occupancy sensing data and classification algorithms to consecutively provide occupancy information for the energy modeling effort.

The Implementation of Optimization Methods for Contrast Enhancement

The performances of the multivariate techniques are directly related to the variable selection process, which is time consuming and requires resources for testing each possible parameter to achieve the best results. Therefore, optimization methods for variable selection process have been proposed in the literature to find the optimal solution in short time by using less system resources. Contrast enhancement is the one of the most important and the parameter dependent image enhancement technique. In this study, two optimization methods are employed for the variable selection for the contrast enhancement technique. Particle swarm optimization (PSO) and artificial bee colony (ABC) optimization methods are implemented to the histogram stretching technique in parameter selection process. The results of the optimized histogram stretching technique are compared with one of the parameter independent contrast enhancement technique; histogram equalization. The results show that the performance of the optimized histogram stretching is better not only in distorted images but also in original images. Histogram equalization degraded the original images while the optimized histogram stretching has no effect due to being an adaptive solution.

An Effective Approach to Input Variable Selection for Preliminary Cost Estimation of Construction Projects

Accurate cost estimates are vital to the effective realisation of construction projects. Extended knowledge, wide‐ranging information, substantial expertise, and continuous improvement are required to attain accurate cost estimation. Cost estimation at the preliminary phase of the project is always a challenge as only limited information is available. Hence, rational selection of input variables for preliminary cost estimation could be imperative. A systematic input variable selection approach for preliminary estimating using an integrated methodology of factor analysis and fuzzy AHP is presented in this paper. First, the factor analysis is used to classify and reduce the input variables and their variable coefficients are determined. Second, fuzzy AHP based on the geometric mean method is employed to determine the weights of input variables in a fuzzy environment where the subjectivity and vagueness are handled with natural language expressions parameterized by triangular fuzzy numbers. Then, the input variables are suggested to be selected starting with those having high coefficient and high importance weight. A set of three variables, one from each group, can be added to the estimating model at a time so that the problem of collinearity can vanish and good accuracy of the estimate can be ensured. The proposed approach enables cost estimators to better understand the complete input variable selection process at the early stage of project development and provide a more accurate, rational, and systematic decision support tool.

Impact of Individual Comorbidities on Treatment Outcomes in Chronic Lymphocytic Leukemia

Introduction: Chronic lymphocytic leukemia (CLL) is a common leukemia which tends to occur late in life. Comorbidities are common, and the iwCLL guidelines recommend their assessment in patients (pts) enrolled on clinical trials. The Cumulative Illness Rating Scale (CIRS) is a rigorous tool designed to evaluate the burden of comorbidities, which has been employed in therapeutic studies. Our group and others demonstrated that CIRS score predicts survival in pts with CLL treated with either chemo-immunotherapy (CIT) or novel kinase inhibitors (KI; ibrutinib) (Manda et al, 2016 & Gordon et al, 2018). However, CIRS has not become part of common clinical practice, in part due to complexities in scoring. It is also unknown whether all of the 14 organ systems included in the score carry equal weight to determine prognosis. Here we report the impact of specific comorbidities from a multicenter retrospective cohort of CLL pts treated with either CIT or KI. Methods: We conducted a retrospective analysis of pts with CLL treated at five US academic medical centers between 2000 and 2017. CIRS score was calculated as in Salvi et al, 2008. Random forest (RF) was used to assess specific comorbidities9 impact on overall survival (OS) and event-free survival (EFS, defined as time to new therapy, disease progression or death). We adapted two separate approaches to investigate the RF variable selection process: variable Importance (VIMP), a property related to variable misspecification, and Minimal Depth (MD), a property derived from the construction of trees within the forest. Best variables were those selected consistently as top 3 in both VIMP and MD on the 500 RF repetitions. Because hepatic and renal comorbidities were rare they were excluded. OS and EFS were assessed by Kaplan-Meier estimates and Cox proportional hazard model adjusted for performance status and age. Significance was assessed with log-rank test. Results: 398 pts were included in the final analysis. The median age was 63 years (range, 30-93). 50% of pts (n=198) had a high CIRS score (≥7). 184 pts (46%) had comorbidities assessed in relapsed setting. For all pts, the most common treatments included ibrutinib (n=145; 37%), fludarabine-containing regimens (n=104; 26%) and bendamustine (n=39; 10%). Complex karyotype was observed in 3.5% (n=14) and 10.6% (n=42) of pts had del(17p). Pts with comorbidities (CIRS ≥7) demonstrated shortened survival following therapy, with 5-year OS of 64% vs 89% (p Random forest variable selections identified vascular comorbidities (e.g. DVT/PE) as the most influential risk factor for OS with CIT treatment, while HEENT and cardiac comorbidities were most impactful to OS for patients treated with KI. For EFS, the most influential comorbidities were cardiac and vascular for the CIT treatment group and endocrine and HEENT for patients treated with KI. Across EFS and OS, the most frequently selected variables in CIT were cardiac, hypertension, vascular and neurologic. We constructed a simplified scoring system assigning 1 point for each category. Comparing scores of 0, 1 and 2-4 (n=100, n=82, n=60), 5-year OS was 87%, 82% and 66%, respectively (p Conclusion: Comorbidities impact survival in CLL whether treated with CIT or KI. Which comorbidities are most prognostic may vary by treatment type. Vascular and cardiac comorbidities appear to be the most relevant in CLL pts treated with CIT. Meanwhile, cardiac, endocrine and HEENT had greater impact when pts were treated with KI. A simplified CIRS score is predictive of outcomes in both treatment subgroups. Disclosures Choi:Gilead: Speakers Bureau; AbbVie, Inc: Consultancy, Speakers Bureau; Pharmacyclics: Consultancy, Research Funding, Speakers Bureau; Rigel: Consultancy; Genentech: Speakers Bureau. Cohen:Takeda: Research Funding; Pharmacyclics: Consultancy, Membership on an entity9s Board of Directors or advisory committees; Infinity Pharmaceuticals: Consultancy, Membership on an entity9s Board of Directors or advisory committees; Millennium: Consultancy, Membership on an entity9s Board of Directors or advisory committees; Seattle Genetics: Consultancy, Membership on an entity9s Board of Directors or advisory committees, Research Funding; Novartis: Consultancy, Membership on an entity9s Board of Directors or advisory committees, Research Funding; Janssen: Research Funding; AbbVie: Consultancy, Membership on an entity9s Board of Directors or advisory committees; Bristol-Myers Squibb: Research Funding; Celgene: Consultancy, Membership on an entity9s Board of Directors or advisory committees; BioInvent: Consultancy. Persky:Genentech: Honoraria; Morphosys (IDMC): Consultancy; Spectrum: Research Funding; Merck: Research Funding. Danilov:Aptose Biosciences: Research Funding; Verastem: Consultancy, Research Funding; Astra Zeneca: Consultancy; Gilead Sciences: Consultancy, Research Funding; Takeda Oncology: Research Funding; Genentech: Consultancy, Research Funding; TG Therapeutics: Consultancy; Bayer Oncology: Consultancy, Research Funding.

A Sub-Model Theorem for Ordinary Least Squares

Variable selection or subset selection is an important step in the process of model fitting. There are many ways to select the best subset of variables including Forward selection, Backward elimination, etcetera. Ordinary least squares (OLS) is one of the most commonly used methods of fitting the final model. Final sub-model can perform poorly if the variable selection process failed to choose the right number of variables. This paper gives a new theorem and a mathematical proof to illustrate the reason for the poor performances, when using the least squares method after variable selection.

Assessing Associations between School Facility Condition, Neighborhood Environment, and Respiratory Health in Public Schools: An Application of Ultrahigh Dimensional Variable Selection Method on Big Data

Background: While poor school facility conditions were associated with impaired academic performance in large-scale studies, less research has assessed its health impact among students. Furthermore, strong correlations between variables made it difficult to assess a wide range of environmental factors in one model. In this study, we aim to assess the impact of school and home environmental factors on student's respiratory health in a multi-pollutant model by applying an innovative variable selection method.Method: A case-control study was conducted using statewide hospital records from patients aged 4 to 18 during 2010-2013 in upstate New York. Emergency department (ED) visits due to respiratory diseases were defined as cases, while visits due to gastrointestinal diseases, injury, and appendicitis were used as controls. Each patient was assigned to a school based on the information on patient's age, home address, and school zones. A set of 128 variables indicating school facility conditions and school or home neighborhood environmental hazards were included in the variable selection process. Iterative sure independence screening was applied to construct the final multi-pollutant regression model.Result: Eighteen environmental factors were chosen for the final model, including 11 indicators of school facility conditions and 7 indicators of traffic/industrial emission around the school/home. The strongest associations were found between respiratory ED visit and conditions of tile floor, foundation type, noise from air-conditioning or traffic, PM2.5 level in home neighborhood, active leak in classroom, and heating fuel system (increased risks ranged from 9% to 17%).Conclusion: Application of advanced variable selection method allows construction of multi-pollutant models with wide range of environmental factors. Specific school facility conditions potentially play a crucial role in student’s respiratory health even after controlling for neighborhood environment.

High-dimensional prediction of binary outcomes in the presence of between-study heterogeneity

Many prediction methods have been proposed in the literature, but most of them ignore heterogeneity between populations. Either only data from a single study or population is available for model building and evaluation, or when data from multiple studies make up the training dataset, studies are pooled before model building. As a result, prediction models might perform less than expected when applied to new subjects from new study populations. We propose a linear method for building prediction models with high-dimensional data from multiple studies. Our method explicitly addresses between-population variability and tends to select predictors that are predictive in most of the study populations. We employ empirical Bayes estimators and hence avoid selection bias during the variable selection process. Simulation results demonstrate that the new method works better than other linear prediction methods that ignore the between-study variability. Our method is developed for classification into two groups.

An ontology-guided semantic data integration framework to support integrative data analysis of cancer survival

BackgroundCancer is the second leading cause of death in the United States, exceeded only by heart disease. Extant cancer survival analyses have primarily focused on individual-level factors due to limited data availability from a single data source. There is a need to integrate data from different sources to simultaneously study as much risk factors as possible. Thus, we proposed an ontology-based approach to integrate heterogeneous datasets addressing key data integration challenges.MethodsFollowing best practices in ontology engineering, we created the Ontology for Cancer Research Variables (OCRV) adapting existing semantic resources such as the National Cancer Institute (NCI) Thesaurus. Using the global-as-view data integration approach, we created mapping axioms to link the data elements in different sources to OCRV. Implemented upon the Ontop platform, we built a data integration pipeline to query, extract, and transform data in relational databases using semantic queries into a pooled dataset according to the downstream multi-level Integrative Data Analysis (IDA) needs.ResultsBased on our use cases in the cancer survival IDA, we created tailored ontological structures in OCRV to facilitate the data integration tasks. Specifically, we created a flexible framework addressing key integration challenges: (1) using a shared, controlled vocabulary to make data understandable to both human and computers, (2) explicitly modeling the semantic relationships makes it possible to compute and reason with the data, (3) linking patients to contextual and environmental factors through geographic variables, (4) being able to document the data manipulation and integration processes clearly in the ontologies.ConclusionsUsing an ontology-based data integration approach not only standardizes the definitions of data variables through a common, controlled vocabulary, but also makes the semantic relationships among variables from different sources explicit and clear to all users of the same datasets. Such an approach resolves the ambiguity in variable selection, extraction and integration processes and thus improve reproducibility of the IDA.

Variable selection - A review and recommendations for the practicing statistician.

Statistical models support medical research by facilitating individualized outcome prognostication conditional on independent variables or by estimating effects of risk factors adjusted for covariates. Theory of statistical models is well‐established if the set of independent variables to consider is fixed and small. Hence, we can assume that effect estimates are unbiased and the usual methods for confidence interval estimation are valid. In routine work, however, it is not known a priori which covariates should be included in a model, and often we are confronted with the number of candidate variables in the range 10–30. This number is often too large to be considered in a statistical model. We provide an overview of various available variable selection methods that are based on significance or information criteria, penalized likelihood, the change‐in‐estimate criterion, background knowledge, or combinations thereof. These methods were usually developed in the context of a linear regression model and then transferred to more generalized linear models or models for censored survival data. Variable selection, in particular if used in explanatory modeling where effect estimates are of central interest, can compromise stability of a final model, unbiasedness of regression coefficients, and validity of p‐values or confidence intervals. Therefore, we give pragmatic recommendations for the practicing statistician on application of variable selection methods in general (low‐dimensional) modeling problems and on performing stability investigations and inference. We also propose some quantities based on resampling the entire variable selection process to be routinely reported by software packages offering automated variable selection algorithms.

SELEÇÃO DE VARIÁVEIS PARA CATEGORIZAÇÃO DE AMOSTRAS QUÍMICAS

This paper presents a method to select the best variables to categorize chemical samples into two classes, say conforming or non-conforming. For that matter, PLS regression is combined with a data mining tool, the k-Nearest Neighbor classification technique, through an iterative variable selection process. The recommended subset of variables is chosen based on several criteria: sensitivity, specificity and percent of retained variables. When applied to two datasets related to wine analysis and one associated to QSAR, the proposed method significantly reduced the number of variables required for classification, while yielding superior categorization performance when compared to using all original variables.

Applying systems thinking to inform studies of wildlife trade in primates.

Wildlife trade presents a major threat to primate populations, which are in demand from local to international scales for a variety of uses from food and traditional medicine to the exotic pet trade. We argue that an interdisciplinary framework to facilitate integration of socioeconomic, anthropological, and biological data across multiple spatial and temporal scales is essential to guide the study of wildlife trade dynamics and its impacts on primate populations. Here, we present a new way to design research on wildlife trade in primates using a systems thinking framework. We discuss how we constructed our framework, which follows a social-ecological system framework, to design an ongoing study of local, regional, and international slow loris (Nycticebus spp.) trade in Vietnam. We outline the process of iterative variable exploration and selection via this framework to inform study design. Our framework, guided by systems thinking, enables recognition of complexity in study design, from which the results can inform more holistic, site-appropriate, and effective trade management practices. We place our framework in the context of other approaches to studying wildlife trade and discuss options to address foreseeable challenges to implementing this new framework.

Tackling concussion in professional rugby union: a case–control study of tackle-based risk factors and recommendations for primary prevention

Background/aimConcussion, the most common injury in professional rugby union, occurs most commonly during the tackle. Thus, we investigated the association between tackle characteristics and concussion.Methods182 video clips of tackles leading...

Application of artificial neural networks for determining energy-efficient operating set-points of the VRF cooling system

The aim of this study was to develop an Artificial Neural Network (ANN) model that can predict the amount of cooling energy consumption for the different settings of the variable refrigerant flow (VRF) cooling system's control variables. Matrix laboratory (MATLAB) and its neural network toolbox were used for the ANN model development and test performance. For the model training and performance evaluation, data sets were collected through the field measurement. Four steps were conducted in the development process: initial model development, input variable selection, model optimization, and performance evaluation. In the initial model development and input variable selection process, seven input variables were selected as input neurons: TEMPOUT, HUMIDOUT, TEMPIN, LOADCOOL, TEMPSA, TEMPCOND, and PRESCOND. In addition, the initial model was optimized to have 2 hidden layers, 15 hidden neurons in each hidden layer, a learning rate of 0.3, and a momentum of 0.3. The optimized model demonstrated its prediction accuracy within the recommended level, thus proved its potential for application in the control algorithm for creating a comfortable indoor thermal environment in an energy-efficient manner.

Modelling residential land values using geographic and geometric accessibility in Guatemala City

Location and accessibility are core concepts for land-value research. However, the perspective is still limited in their conceptual and methodological application to cities from the Global South. The objective of this research is to bridge concepts and definitions to comprehensively operationalize accessibility indicators and uncover its relation with residential land-values in Guatemala City. We developed a multivariate regression model using the following access metrics: (1) geographic-access indices that were computed using time-based analyses per transport mode; (2) geometric-access metrics estimated via Space Syntax at various spatial scales; (3) a proposed geometric via geographic-access metric computed as potential access to network centrality. A variable selection process allowed to assess the information contribution of each variable in building a parsimonious model. We assessed the model in the context of model variations that represent common approaches used in existing literature. Geographic access to the core business district has the highest impact on the land-values, followed by proximity to urban areas with high geometric-access, measured as geometric via geographic access. Geometric accessibility at neighbourhood and city-wide scales add spatialized information that contributes to a parsimonious model and reduces spatial dependence. The model yielded the highest goodness of fit and prediction accuracy compared with the model variations. We concluded that Guatemala City land-values follow a predominant monocentric structure. Additionally, potential access to vital urban areas as identified via Space Syntax denotes the presence of economic activities, or potential for such, which were not explicitly addressed through the geographic-access metrics. The results have limitations but pose methodological possibilities relevant for research and practice in similar Latin American cities.

Assessing the accuracy and stability of variable selection methods for random forest modeling in ecology.

Random forest (RF) modeling has emerged as an important statistical learning method in ecology due to its exceptional predictive performance. However, for large and complex ecological data sets, there is limited guidance on variable selection methods for RF modeling. Typically, either a preselected set of predictor variables are used or stepwise procedures are employed which iteratively remove variables according to their importance measures. This paper investigates the application of variable selection methods to RF models for predicting probable biological stream condition. Our motivating data set consists of the good/poor condition of n = 1365 stream survey sites from the 2008/2009 National Rivers and Stream Assessment, and a large set (p = 212) of landscape features from the StreamCat data set as potential predictors. We compare two types of RF models: a full variable set model with all 212 predictors and a reduced variable set model selected using a backward elimination approach. We assess model accuracy using RF's internal out-of-bag estimate, and a cross-validation procedure with validation folds external to the variable selection process. We also assess the stability of the spatial predictions generated by the RF models to changes in the number of predictors and argue that model selection needs to consider both accuracy and stability. The results suggest that RF modeling is robust to the inclusion of many variables of moderate to low importance. We found no substantial improvement in cross-validated accuracy as a result of variable reduction. Moreover, the backward elimination procedure tended to select too few variables and exhibited numerous issues such as upwardly biased out-of-bag accuracy estimates and instabilities in the spatial predictions. We use simulations to further support and generalize results from the analysis of real data. A main purpose of this work is to elucidate issues of model selection bias and instability to ecologists interested in using RF to develop predictive models with large environmental data sets.