Variable Screening Methods Research Articles

A two-stage ensemble learning based prediction and grading model for PD-1/PD-L1 inhibitor-related cardiac adverse events: A multicenter retrospective study

BackgroundImmune-related cardiac adverse events (ircAEs) caused by programmed cell death protein-1 (PD-1) and programmed death-ligand-1 (PD-L1) inhibitors can lead to fulminant and even fatal consequences. This study aims to develop a prediction and grading model for ircAEs, enabling graded management of patients. MethodsThis study utilized medical record systems from two medical institutions to develop a prediction and grading model for ircAEs using ten machine learning algorithms and two variable screening methods. The model was developed based on a two-stage ensemble learning framework. In the first stage, the ircAEs and non-ircAEs cases were classified. In the second stage, ircAEs cases were grouped into grades 1–2 and 3–5. The experiments were evaluated using five-fold cross-validation. The model's prediction performance was assessed using accuracy, precision, recall, F1 value, Brier score, receiver operating characteristic curve area (AUC), and area under the precision-recall curve (AUPR). Results615 patients were included in the study. 147 experienced ircAEs, and 44 experienced grade 3–5 ircAEs. The soft voting classifier trained using the variables screened by feature importance ranking performed better than other classifiers in both stages. The average AUC for the first and second stages is 84.18 % and 85.13 %, respectively. In the first stage, the three most important variables are N-terminal B-type natriuretic peptide (NT-proBNP), interleukin-2 (IL-2), and C-reactive protein (CRP). In the second stage, the patient's age, NT-proBNP, and left ventricular ejection fraction (LVEF) are the three most critical variables. ConclusionsThe prediction and grading model of ircAEs based on two-stage ensemble learning established in this study has good performance and potential clinical application.

Comparison of QRNN and QRF Models in Forest Biomass Estimation Based on the Screening of VIs Using an Equidistant Quantile Method

The investigation of a potential correlation between the filtered-out vegetation index and forest aboveground biomass (AGB) using the conventional variables screening method is crucial for enhancing the estimation accuracy. In this study, we examined the Pinus densata forests in Shangri-La and utilized 31 variables to establish quantile regression models for the AGB across 19 quantiles. The key variables associated with biomass were based on their significant correlation with the AGB in different quantiles, and the QRNN and QRF models were constructed accordingly. Furthermore, the optimal quartile models yielding the minimum mean error were combined as the best QRF (QRFb) and QRNN (QRNNb). The results were as follows: (1) certain bands exhibited significant relationships with the AGB in specific quantiles, highlighting the importance of band selection. (2) The vegetation index involving the band of blue and SWIR was more suitable for estimating the Pinus densata. (3) Both the QRNN and QRF models demonstrated their optimal performance in the 0.5 quantiles, with respective R2 values of 0.68 and 0.7. Moreover, the QRNNb achieved a high R2 value of 0.93, while the QRFb attained an R2 value of 0.86, effectively reducing the underestimation and overestimation. Overall, this research provides valuable insights into the variable screening methods that enhance estimation accuracy and mitigate underestimation and overestimation issues.

Tracing the geographic origin of CTC black tea based on colorimetric sensor array response to taste substances combined with chemometrics

The quality of crush–tear–curl black tea (CTC-BT) varies greatly by geographic origin. Origin traceability is crucial for consumer interest protection, market order regulation, and food safety monitoring. This paper proposes a fast and accurate method for qualitative discrimination of CTC-BT origins and quantitative detection of its key taste-presenting substances. The method involves a simple colorimetric sensor array and ultraviolet–visible spectroscopy. The effects of various variable screening methods on modeling results were compared. A particle swarm optimization–based support vector machine achieved the highest performance in qualitative discrimination, with a correct classification rate of 99.48%. Based on their origin-distinguishing contributions and dose-over-thresholds, seven key taste-presenting substances were screened, namely, theaflavin, caffine, vitexin-2-O-rhamnoside, rutin, epigallocatechin gallate, epicatechin gallate, gallic acid. A least squares-support vector regression model achieved accurate quantification of the seven aforementioned compounds (square root of determination coefficient of prediction >0.9698, residual prediction deviation >2).

E-nose and colorimetric sensor array combining homologous data fusion strategy discriminating the roasting degree of large-leaf yellow tea

Different degrees of roasting result in differences in the quality and flavor of large-leaf yellow tea. The current sensory evaluation and chemical detection methods cannot meet the requirement of online differentiation of LYT roasting degree, so an accurate and comprehensive assessment method needs to be developed urgently. First, the two aroma sensing technologies were compared. Two variable screening methods and three recognition algorithms were employed to build discriminant models. The results showed that the discrimination rate of the colorimetric sensor array (CSA) in the prediction set reached 91.89 %, outperforming that of the E-nose. Subsequently, three fusion strategies were applied to improve the discrimination accuracy. The discrimination rate of the middle fusion strategy resulted in an optimal resolution of 94.59 %. The results obtained from the homologous fusion were able to evaluate the roasting degree comprehensively and accurately, which provides a new method and idea for tea aroma quality.

Assessing the Current and Future Potential Distribution of Solanum rostratum Dunal in China Using Multisource Remote Sensing Data and Principal Component Analysis

Accurate information concerning the spatial distribution of invasive alien species’ habitats is essential for invasive species prevention and management, and ecological sustainability. Currently, nationwide identification of suitable habitats for the highly destructive and potentially invasive weed, Solanum rostratum Dunal (S. rostratum), poses a series of challenges. Simultaneously, research on potential future invasion areas and likely directions of spread has not received adequate attention. This study, based on species occurrence data and multi-dimensional environmental variables constructed from multi-source remote sensing data, utilized Principal Component Analysis (PCA) in combination with the Maxent model to effectively model the current and future potential habitat distribution of S. rostratum in China, while quantitatively assessing the various factors influencing its distribution. Research findings indicate that the current suitable habitat area of S. rostratum covers 1.3952 million km2, all of which is located in northern China. As the trend of climate warming persists, the potential habitat suitability range of S. rostratum is projected to shift southward and expand in the future; while still predominantly located in northern China, it will have varying degrees of expansion at different time frames. Notably, during the period from 2040 to 2061, under the SSP1-2.6 scenario, the habitat area exhibits the most significant increase, surpassing the current scenario by 19.23%. Furthermore, attribution analysis based on PCA inverse transformation reveals that a combination of soil, climate, spatial, humanistic, and topographic variables collectively influence the suitability of S. rostratum habitats, with soil factors, in particular, playing a dominant role and contributing up to 75.85%. This study identifies target areas for the management and control of S. rostratum, providing valuable insights into factor selection and variable screening methods in species distribution modeling (SDM).

NIR Spectral Inversion of Soil Physicochemical Properties in Tea Plantations under Different Particle Size States.

Soil fertility is vital for the growth of tea plants. The physicochemical properties of soil play a key role in the evaluation of soil fertility. Thus, realizing the rapid and accurate detection of soil physicochemical properties is of great significance for promoting the development of precision agriculture in tea plantations. In recent years, spectral data have become an important tool for the non-destructive testing of soil physicochemical properties. In this study, a support vector regression (SVR) model was constructed to model the hydrolyzed nitrogen, available potassium, and effective phosphorus in tea plantation soils of different grain sizes. Then, the successful projections algorithm (SPA) and least-angle regression (LAR) and bootstrapping soft shrinkage (BOSS) variable importance screening methods were used to optimize the variables in the soil physicochemical properties. The findings demonstrated that soil particle sizes of 0.25-0.5 mm produced the best predictions for all three physicochemical properties. After further using the dimensionality reduction approach, the LAR algorithm (R2C = 0.979, R2P = 0.976, RPD = 6.613) performed optimally in the prediction model for hydrolytic nitrogen at a soil particle size of 0.25~0.5. The models using data dimensionality reduction and those that used the BOSS method to estimate available potassium (R2C = 0.977, R2P = 0.981, RPD = 7.222) and effective phosphorus (R2C = 0.969, R2P = 0.964, RPD = 5.163) had the best accuracy. In order to offer a reference for the accurate detection of soil physicochemical properties in tea plantations, this study investigated the modeling effect of each physicochemical property under various soil particle sizes and integrated the regression model with various downscaling strategies.

Estimation of Soil Salt Content at Different Depths Using UAV Multi-Spectral Remote Sensing Combined with Machine Learning Algorithms

Soil salinization seriously affects the sustainable development of agricultural production; thus, the timely, efficient, and accurate estimation of soil salt content (SSC) has important research significance. In this study, the feasibility of soil salt content retrieval using machine learning models was explored based on a UAV (unmanned aerial vehicle) multi-spectral remote sensing platform. First, two variable screening methods (Pearson correlation analysis and Grey relational analysis) are used to screen the characteristic importance of 20 commonly used spectral indices. Then, the sensitive spectral variables were divided into a vegetation index group, a salt index group, and a combination variable group, which represent the model. To estimate SSC information for soil depths of 0–20 cm and 20–40 cm, three machine learning regression models were constructed: Support Vector Machine (SVM), Random Forest (RF), and Backpropagation Neural Network (BPNN). Finally, the salt distribution map for a 0–20 cm soil depth was drawn based on the best estimation model. The results of experiments show that GRA is better than PCA in improving the accuracy of the estimation model, and the combination variable group containing soil moisture information performs best. The three machine learning models have achieved good prediction effects to some extent. The accuracy and stability of the model are considered comprehensively, the prediction effect of 0–20 cm is higher than that of 20–40 cm, and the validation set coefficient of determination (R2), Root-Mean-Square-Error (RMSE), and Mean Absolute Error (MAE) of the best inversion model are 0.775, 0.055, and 0.038, and the soil salt spatial map based on the optimal estimation model can reflect the salinization distribution in the study area. Therefore, this study shows that a UAV multi-spectral remote sensing platform combined with machine learning models can better monitor farmland soil salt content.

Variable screening methods in spatial infectious disease transmission models

Data-driven mathematical modelling can enrich our understanding of infectious disease spread enormously. Individual-level models of infectious disease transmission allow the incorporation of different individual-level covariates, such as spatial location, vaccination status, etc. This study aims to explore and develop methods for fitting such models when we have many potential covariates to include in the model. The aim is to enhance the performance and interpretability of models and ease the computational burden of fitting these models to data. We have applied and compared multiple variable selection methods in the context of spatial epidemic data. These include a Bayesian two-stage least absolute shrinkage and selection operator (Lasso), forward and backward stepwise selection based on the Akaike information criterion (AIC), spike-and-slab priors, and random variable selection (boosting) methods. We discuss and compare the performance of these methods via simulated datasets and UK 2001 foot-and-mouth disease data. While comparing the variable selection methods all performed consistently well except the two-stage Lasso. We conclude that the spike-and-slab prior method is to be recommended, consistently resulting in high accuracy and short computational time.

Overview of High-Dimensional Measurement Error Regression Models

High-dimensional measurement error data are becoming more prevalent across various fields. Research on measurement error regression models has gained momentum due to the risk of drawing inaccurate conclusions if measurement errors are ignored. When the dimension p is larger than the sample size n, it is challenging to develop statistical inference methods for high-dimensional measurement error regression models due to the existence of bias, nonconvexity of the objective function, high computational cost and many other difficulties. Over the past few years, some works have overcome the aforementioned difficulties and proposed several novel statistical inference methods. This paper mainly reviews the current development on estimation, hypothesis testing and variable screening methods for high-dimensional measurement error regression models and shows the theoretical results of these methods with some directions worthy of exploring in future research.

A screening method for ultra-high dimensional features with overlapped partition structures.

Ultra-high dimensional data, such as gene and neuroimaging data, are becoming increasingly important in biomedical science. Identifying important biomarkers from the huge number of features can help us gain better insights into further researches. Variable screening is an efficient tool to achieve this goal under the large scale cases, which reduces the dimension of features into a moderate size by removing the major part of inactive ones. Developing novel variable screening methods for high-dimensional features with group structures is challenging, especially under the overlapped cases. For example, the huge-scaled genes usually can be partitioned into hundreds of pathways according to background knowledge. One primary characteristic for this type of data is that many genes may appear across more than one pathway, which means that different pathways are overlapped. However, existing variable screening methods only could deal with disjoint group structure cases. To fill this gap, we propose a novel variable screening method for the generalized linear model by incorporating overlapped partition structures with theoretical guarantee. Besides the sure screening property, we also test the performance of the proposed method through a series of numerical studies and apply it to statistical analysis of a breast cancer data.

Non-destructive assessment of equivalent umami concentrations in salmon using hyperspectral imaging technology combined with multivariate algorithms

This study utilized equivalent umami concentrations (EUC) to characterize umami intensity in salmon with different freeze–thaw times. A rapid and non-destructive method was established to determine EUC values in salmon which is based on hyperspectral imaging (HSI) system combined with multiple characteristic variable screening methods. The established CARS-PLS model showed greater advantages in correlating the reference values of spectral data with EUC in salmon with Rc of 0.9012, Rp of 0.9009, RMSECV of 0.82, and RMSEP of 0.88. The model was employed pixel-wise to visualize the distribution of EUC with different freeze–thaw times, which demonstrated the reduction of EUC value with the increasing of freeze–thaw times. Therefore, this reseearch showed hyperspectral imaging (HSI) system combined with chemometrics possesses a substantial capability to predict and visualize the EUC of salmon, which would provide an intuitive understanding of salmon quality prediction and detection.

Rapid detection of multiple colorant adulteration in Keemun black tea based on hemp spherical AgNPs-SERS

This study synthesized stable and sensitive hemp spherical AgNPs as the SERS substrate for the simultaneous and rapid detection of sunset yellow, lemon yellow, carmine and erythrosine adulteration in black tea. With R6G as the probe molecule, the AgNPs were determined to have satisfactory stability over 60 days with an enhancement factor of 108. The effects of three variable screening methods on model performance were compared. Among them, CARS-PLS exhibited superior performance for the quantification of all the four colorants, with prediction set correlation coefficients of 0.95, 0.97, 0.99 and 0.88, respectively. The differentiation of the mixed colorants was also achieved, with recoveries ranging from 91.87 % to 106.5 % with RSD value <1.97 %, demonstrating the high accuracy and precision of the proposed method. The results indicate that AgNPs-based SERS is an effective method and has substantial potential for application in the identification and quantification of colorant in tea.

UAV-based chlorophyll content estimation by evaluating vegetation index responses under different crop coverages

• LCC was estimated based on the UAV-based image analysis. • The response stability of vegetation index and LCC was analyzed under different coverage. • Different variable screening methods were used to optimize the spectral parameters. • The spatial distribution map of maize LCC was constructed to provide potentials for fertilization application. Efficiently estimating chlorophyll content is important in monitoring the photosynthesis capacity and growth status of maize canopy in precision agriculture management. Vegetation index (VI) easily obtained by proximal remote sensing has been used as a non-destructive and high-throughput way in crop monitoring, especially in chlorophyll estimation. However, the estimated results of the field chlorophyll content by VIs always face challenges from soil background inhibition and estimation stability under the dynamic changes of vegetation biomass. Thus, an unmanned aerial vehicle (UAV)-based chlorophyll content estimation was conducted by evaluating VI responses under different crop coverages. An analysis was conducted on 36 VIs under different crop coverage conditions to explore their response differences and robustness for chlorophyll estimation. This work focused on the three kinds of VIs named simple vegetation index, modified vegetation index, and functional vegetation index. In 2020, at the experimental station of Dryland Farming Institute of Hebei Academy of Agriculture and Forestry Sciences, UAV carrying multispectral sensor was used to collect visible and near-infrared images of the canopy at the jointing stage of maize under six fertilization levels to obtain VIs. After the UAV fled, ground calibration and sample collection were performed simultaneously, and chlorophyll content was measured. For data processing, correlation coefficient method (CCM) and maximal information coefficient (MIC) were first used to analyze the correlation response characteristics of VIs and chlorophyll content under three different coverage levels. The results showed that when the level of canopy coverage was increased, the linear correlation between VIs and chlorophyll content was substantially reduced. The MIC response indicating linear and non-linear combination relationship was more robust. In addition, the VIs obtained by UAV had a significant linear correlation with maize canopy chlorophyll under low (0.05–0.35) and medium (0.35–0.48) coverage, but an obvious non-linear correlation under high (0.48–0.75) coverage. Chlorophyll-sensitive parameters were then screened based on methods of CCM, MIC, and random frog method (RFM), respectively. Partial least squares regression (PLS) and random forest (RF) algorithms were used to establish the maize canopy chlorophyll content detection models. The findings showed that when Green minus red vegetation index (GMR), Red light normalized value (NRI), Normalized difference red edge (NDRE), Modified simple ratio with red edge (MSRREG), Enhanced Vegetation Index (EVI), Normalized red green difference vegetation index (NDIg), Normalized red blue difference vegetation index (NDIb), Soil-adjusted vegetation index (SAVI), Optimized soil-adjusted vegetation index with red edge (OSAVIREG), Soil-atmospherically resistant vegetation index (SARVI) were selected based on RFM as the optimal spectral variables, the chlorophyll content detection model constructed based on PLS had the least numbers of characteristic variables and the best model accuracy. The training set R 2 and RMSE were 0.753 and 2.089 mg/L, respectively, and the verification set R 2 and RMSE were 0.682 and 2.361 mg/L, respectively. Field chlorophyll content and detection error distribution maps were also drawn and combined with the distribution of fertilization management to provide support for the UAV monitoring of crop growth in the field and variable fertilization management decisions.

Improved Na+ estimation from hyperspectral data of saline vegetation by machine learning

Monitoring the growth state of vegetation using remote sensing is the current trends in agricultural research. This study aims to identify an optimal hyperspectral vegetation extraction framework to improve leaf Na+ monitoring in the northwestern part of China based on the hyperspectral data of saline vegetation. The Partial Least Squares (PLS), Support Vector Machine (SVM), Random Forest (RF) models were constructed to model the leaf Na+, while the Aggregated Boosted Tree (ABT) and Random Forest (RF) variable importance screening methods were used to optimize the variables in the leaf Na+ extraction. Then, the optimal variable screening method and the model of inverting vegetation Na+ was identified. The results showed that the estimation of Na+ content within saline vegetation leaves by constructing spectral indices is feasible as 33 vegetation indices meets the requirements, the RF (R2 = 0.73, RMSE = 0.50) and PLS (R2 = 0.72, RMSE = 0.59) models are relatively good, followed by the SVM (R2 = 0.68, RMSE = 0.53) model. In addition, all the three models have been improved using the ABT variable importance screening method, where the RF (R2 = 0.81, RMSE = 0.42) model had the most satisfactory effect. Similarly, based on the RF importance screening method, all the three models have improved significantly, among which the most effective was the SVM (R2 = 0.82, RMSE = 0.45) model. This study indicates that ABT-RF and RF-SVM are the most ideal combination framework to invert the Na+ content of saline vegetation leaves. This study brings out some inspiration for the combination between the screening approach of variables and model building, improving the accuracy of hyperspectral sensor to monitor the changes in the relevant chemical characteristics of vegetation.

RaSE: A Variable Screening Framework via Random Subspace Ensembles

Variable screening methods have been shown to be effective in dimension reduction under the ultra-high dimensional setting. Most existing screening methods are designed to rank the predictors according to their individual contributions to the response. As a result, variables that are marginally independent but jointly dependent with the response could be missed. In this work, we propose a new framework for variable screening, random subspace ensemble (RaSE), which works by evaluating the quality of random subspaces that may cover multiple predictors. This new screening framework can be naturally combined with any subspace evaluation criterion, which leads to an array of screening methods. The framework is capable to identify signals with no marginal effect or with high-order interaction effects. It is shown to enjoy the sure screening property and rank consistency. We also develop an iterative version of RaSE screening with theoretical support. Extensive simulation studies and real-data analysis show the effectiveness of the new screening framework.

Targeting Predictors Via Partial Distance Correlation With Applications to Financial Forecasting

High-dimensional time series datasets are becoming increasingly common in various fields of economics and finance. Given the ubiquity of time series data, it is crucial to develop efficient variable screening methods that use the unique features of time series. This article introduces several model-free screening methods based on partial distance correlation and developed specifically to deal with time-dependent data. Methods are developed both for univariate models, such as nonlinear autoregressive models with exogenous predictors (NARX), and multivariate models such as linear or nonlinear VAR models. Sure screening properties are proved for our methods, which depend on the moment conditions, and the strength of dependence in the response and covariate processes, amongst other factors. We show the effectiveness of our methods via extensive simulation studies and an application on forecasting U.S. market returns.

Multi-objective lightweight and crashworthiness collaborative optimisation of commercial vehicle cab

A collaborative optimisation method for the lightweight and crashworthiness of the cab structure is proposed using mixed variable screening, approximate models, multi-objective optimisation and ideal point methods. First, modal and stiffness analyses of the cab model are built. Three impact simulation model of cab are established and verified via ECE R29-02 regulation tests. Subsequently, front impact, front pillar impact, and roof strength model are established following ECE R29-03, which is more stringent than the ECE R29-02. However, the result of the front pillar impact of the cab does not meet the requirements of the ECE R29-03. Several improved A-pillar structures are proposed, among which reinforced rib structure can effectively enhance the occupant’s living space. Then, 18 thickness and 4 material variables are selected as design variables based on the mixed variable screening method. Different approximate models are established for performance responses. The results show that the second-order response surface model has high fitting accuracy in the cab modal and bending and torsional stiffness performance, while the Kriging model has high fitting accuracy in the crashworthiness indicators of the cab. Finally, the multi-objective lightweight and crashworthiness collaborative optimisation of the cab are conducted. Compared with the initial model, the crashworthiness of the optimal cab model is improved and the mass is reduced by 27.1 kg.

Rapid detection of catechins during black tea fermentation based on electrical properties and chemometrics

Catechins are important to evaluate the quality of black tea. In this study, a quantitative prediction model was established based on the measurement of electrical properties and a chemometrics method to detect the catechin content in fermented black tea. The effects of different preprocessing, variable screening methods, and nonlinear algorithms on the model were studied. Results show that the electrical parameters most sensitive to catechin content are equivalent parallel capacitance, loss factor, and reactance mainly at low frequencies (0.05–0.1 kHz). Normalization processing (Zscore), variable combinations' population analysis and the iterative retained information variable algorithm (VCPA-IRIV), and the nonlinear intelligent algorithm random forest (RF) can all effectively improve the performance of a catechin prediction model. In the VCPA-IRIV-RF model, the number of introduced variables was reduced from 162 to 9 with a compression ratio of 94.5%; the root mean square error of prediction and the root mean square error of validation of this model were only 0.269 and 0.214, respectively. The predictive correlation coefficient, correlation coefficient of calibration, and residual predictive deviation increased to 0.988, 0.994, and 5.47, respectively, indicating the good performance of the model. The rapid and nondestructive determination of catechin content in black tea fermentation using a method to detect the electrical properties seems to be practical.

Research on air infiltration predictive models for residential building at different pressure

The pressure difference in buildings under natural state is usually below 10 Pa, and the air change rate at 50 Pa (ACH50) is often used to evaluate building airtightness. There is a dearth of research on air infiltration predictive model at different pressures in China. Moreover, the airflow coefficient (C), a key parameter for air infiltration, is necessary to determine ACH50. Based on prior experimental data, several methods including ordinary least squares (OLS), stepwise regression, partial least squares (PLS) and nonlinear fitting with independent variable screening methods, were employed to establish an airflow coefficient model. The determination coefficient (R2) and the variation coefficient of the root-mean-square error (CV(RMSE)) of these models were compared. The simulation results show that R2 of the airflow coefficient models for apartments and villas increased by a maximum of 25.9% and 2.3%, respectively, using PLS method. The improvement with nonlinear fitting was weaker. Based on K-P model, a conversion model between ACH50 and ACH4 was developed as an air infiltration predictive model under natural state. Blower door and tracer gas tests were conducted to verify the conversion model. The expected error was approximately 10%, which may be caused by measurement errors and shielding from surrounding obstructions. Further studies need to focus on obtaining more experimental data for building airtightness and developing a conversion model for high-rise residential buildings.

Consistent Estimation of Generalized Linear Models with High Dimensional Predictors via Stepwise Regression.

Predictive models play a central role in decision making. Penalized regression approaches, such as least absolute shrinkage and selection operator (LASSO), have been widely used to construct predictive models and explain the impacts of the selected predictors, but the estimates are typically biased. Moreover, when data are ultrahigh-dimensional, penalized regression is usable only after applying variable screening methods to downsize variables. We propose a stepwise procedure for fitting generalized linear models with ultrahigh dimensional predictors. Our procedure can provide a final model; control both false negatives and false positives; and yield consistent estimates, which are useful to gauge the actual effect size of risk factors. Simulations and applications to two clinical studies verify the utility of the method.