An Efficient Elastic Net with Regression Coefficients Method for Variable Selection of Spectrum Data

Abstract

Using the spectrum data for quality prediction always suffers from noise and colinearity, so variable selection method plays an important role to deal with spectrum data. An efficient elastic net with regression coefficients method (Enet-BETA) is proposed to select the significant variables of the spectrum data in this paper. The proposed Enet-BETA method can not only select important variables to make the quality easy to interpret, but also can improve the stability and feasibility of the built model. Enet-BETA method is not prone to overfitting because of the reduction of redundant variables realized by elastic net method. Hypothesis testing is used to further simplify the model and provide a better insight into the nature of process. The experimental results prove that the proposed Enet-BETA method outperforms the other methods in terms of prediction performance and model interpretation.

Universal method for robust detection of circadian state from gene expression

“…It has been shown that PLSR tends to shrink the low-variance predictors but can also inflate the high-variance predictors, making it unstable and yielding a higher prediction error than penalized regression ( 46 – 48 ). In high dimensions, when the number of predictors is much greater than the sample size, elastic net has been shown to yield more accurate results than PLSR ( 48 – 50 ). TimeSignature’s ability to achieve high accuracy using only two samples from an individual with great flexibility in timing (8–12 h apart) using a complement of only 40 genes is a significant advance.…”

Combining Fractional Order Derivative and Spectral Variable Selection for Organic Matter Estimation of Homogeneous Soil Samples by VIS–NIR Spectroscopy

“…ENET shrinks the coefficients of redundant spectral variables to zero, by combining L 1 -norm penalty (lasso) and L 2 -norm penalty (ridge) together [27,28], and then the nonzero spectral variables were considered as effective wavelength variables. Compared with PLS with regression coefficient, elastic net with regression coefficient (named ENET) is more stable and reliable.…”

“…CARS method generally can obtain satisfactory model accuracy and has become a common method in variable selection techniques [17,[23][24][25][26]. In contrast to partial least squares (PLS) with regression coefficient, elastic net (ENET) with regression coefficient represents a novel variable selection technique, which belongs to the method of regression coefficients of the first group [27,28]. Unlike the PLS probably containing the uninformative variables, ENET can effectively shrink the model coefficients of redundant variables to zero.…”

Modeling of Biological Data Based on Regression Methods

“…Lasso penalty expects many coefficients close to zero, and only a small subset to be larger and none zero [5]. Whether 147 there are a group of variables with high multicollinearity, lasso will select only one variable without caring which one is selected [6]. Elastic net simultaneously selects the variables automatically with continuous shrinkage and has the property to select groups of correlated variables.…”

“…It shrinks the regression coefficients by combining L1-norm penalty and L2-norm penalty together. The elastic net identify a higher number of correctly influential variables than the lasso technique, and has lower false positive rate than ridge regression [6,7]. The instability of the lasso regression technique when independent variables are highly correlated as in SNPs in high linkage disequilibrium is overcome using the elastic net that was proposed for analyzing high dimensional data [5].…”