Uninformative Variable Elimination Algorithm Research Articles

Hyperspectral imaging-based detection of soluble solids content of loquat from a small sample

The flavor and market value of loquats are substantially affected by their soluble solids content (SSC). This study explored the ability of hyperspectral imaging combined with multiple regression models to detect SSC of loquats using small sample size. In this study, one hundred and fifteen loquats were collected as samples, and regions of interest in the hyperspectral images of the samples were extracted. The spectral data processed through direct orthogonal signal correction (DOSC), the best among the five processing methods examined, was selected for the feature spectrum process. Successive projection algorithm (SPA), competitive adaptive reweighted sampling (CARS), improved uninformative variable elimination algorithm (imUVE), and imUVE-SPA algorithms were used to extract feature spectra and establish partial least squares regression (PLSR), support vector machine regression (SVR), back-propagation neural network (BPNN), and customized convolutional neural network (CNN) models, respectively. The results demonstrated that the CNN model based on the 18 feature spectra extracted by CARS was the best (correlation coefficient of validation = 0.904, root mean square error of validation = 0.336 %, and residual predictive deviation = 2.339). Consequently, hyperspectral images with small sample size combined with appropriate regression models can be used for nondestructive detection of loquat brix. Furthermore, the wavelength feature extraction algorithm effectively improves the performance of CNN models with small sample sizes. This study can serve as a valuable reference for the nondestructive detection of fruit brix in other limited sample situations.

Low temperature response index for monitoring freezing injury of tea plant.

Freezing damage has been a common natural disaster for tea plantations. Quantitative detection of low temperature stress is significant for evaluating the degree of freezing injury to tea plants. Traditionally, the determination of physicochemical parameters of tea leaves and the investigation of freezing damage phenotype are the main approaches to detect the low temperature stress. However, these methods are time-consuming and laborious. In this study, different low temperature treatments were carried out on tea plants. The low temperature response index (LTRI) was established by measuring seven low temperature-induced components of tea leaves. The hyperspectral data of tea leaves was obtained by hyperspectral imaging and the feature bands were screened by successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS) and uninformative variable elimination (UVE). The LTRI and seven indexes of tea plant were modeled by partial least squares (PLS), support vector machine (SVM), random forests (RF), back propagation (BP) machine learning methods and convolutional neural networks (CNN), long short-term memory (LSTM) deep learning methods. The results indicated that: (1) the best prediction model for the seven indicators was LTRI-UVE-CNN (R2 = 0.890, RMSEP=0.325, RPD=2.904); (2) the feature bands screened by UVE algorithm were more abundant, and the later modeling effect was better than CARS and SPA algorithm; (3) comparing the effects of the six modeling algorithms, the overall modeling effect of the CNN model was better than other models. It can be concluded that out of all the combined models in this paper, the LTRI-UVE-CNN was a promising model for predicting the degree of low temperature stress in tea plants.

Ridge regression coupled with a new uninformative variable elimination algorithm as a new descriptor screening method: Application of data reduction in QSAR study of some sulfonated derivatives as c-Met inhibitors

An uninformative variable elimination algorithm was combined with the Ridge regression method. This combination makes the penalized Ridge method, which is essentially incapable of deleting or selecting a variable, an efficient descriptor screening method for screening of descriptors in QSAR studies. The importance of descriptors was used to identify uninformative and redundant descriptors. The descriptor importance was defined as the ratio of the mean coefficients to the standard deviation of the coefficients derived from the leave-one-out (LOO) Ridge models. Then, with an iterative algorithm, among 392 molecular descriptors obtained after initial preprocessing, 358 uninformative descriptors were identified and removed from the data set. The remaining 34 descriptors were used for further variable selection and modeling through the smoothly clipped absolute deviation (SCAD) method. The proposed screening method was used to screen and reduce the size of QSAR data containing 63 compounds as c-Met inhibitors with anti-cancer activity. For comparison, the SCAD models were constructed using original high-dimensional data (SCAD), and the reduced-size data via the UVE-Ridge screening method (UVE-Ridge-SCAD). Both models were used to predict biological activities (pIC50) for compounds in the external test set. The coefficients of determination (R2) and the FIT parameter for the test data predicted using the SCAD model (with 12 selected descriptors) were equal to 0.8042 and 3.78, respectively. The UVE-Ridge-SCAD model (with 9 descriptors) showed R2 and FIT values of 0.9025 and 4.38, respectively. Higher R2 and FIT values indicate the excellent predictive power of the UVE-Ridge-SCAD model. Finally, the UVE-Ridge-SCAD model was used to suggest new high-potency compounds. The accuracy of the proposed compounds was investigated using ligand-receptor interactions.

Vigor identification of maize seeds by using hyperspectral imaging combined with multivariate data analysis

As an important food crop in China, rapid and accurate discrimination of maize seed vigor is crucial for agricultural development. In this study, 1,680 maize seeds were subjected to artificial accelerated aging treatment with “Zhengdan 958″ as the research object to verify the difference in vigor between damaged and healthy seeds based on the results of the standard germination tests. A Hyperspectral imaging (HSI) system was used to obtain spectral information of samples, and the original spectra were preprocessed using second-order Savitzky-Golay smoothing (SG-2), first derivative (FD), detrending (DE), standard normal variate (SNV), and multiplicative scatter correction (MSC) methods. The 49, 60, 54, and 40 numbers of feature wavelengths (nm) were extracted from the processed spectra using successive projections algorithm (SPA), uninformative variable elimination (UVE), interval random frog (IRF), and iteratively variable subset optimization (IVSO), respectively. The decision tree (DT), support vector machine (SVM), K-nearest neighbor (KNN), linear discriminant analysis (LDA), random forest (RF), and artificial neural network (ANN) models were constructed based on full wavelength and feature wavelength, of which the best model was DE-UVE-ANN, and its identification accuracy reached 95.24 %. The experimental results show that the UVE algorithm is the most effective method for preprocessing, and the accuracy of LDA and ANN models built based on it is above 85.71 % and 89.76 %, respectively. In addition, the hyperspectral images were visualized based on an object-oriented approach to observe intuitive identification results. The results indicate that the proposed method is instructive for the vigor identification of maize seeds.

Investigation of soil properties using different techniques of mid‐infrared spectroscopy

SummaryDiffuse reflectance spectroscopy (DRF), attenuated total reflectance spectroscopy (ATR) and Fourier transform infrared photoacoustic spectroscopy (PAS) are useful techniques that are commonly used to characterize soil properties. Here, these techniques were used to evaluate individual properties of 1456 paddy soil samples from Nanjing, China, to explore their features in the mid‐infrared range. The diffuse reflectance spectroscopy showed broad and shouldered peaks that ranged from 4000 to 500 cm−1. Attenuated total reflectance spectroscopy produced main peaks around 1000 cm−1. Fourier transform infrared photoacoustic spectroscopy spectra showed several moderate intensity peaks in the regions of 4000–2600 and 2500–500 cm−1. The effective variables (wavenumbers) were selected using an uninformative variable elimination (UVE) algorithm to predict soil pH, and soil organic matter (SOM), total nitrogen (TN) and available phosphorus (AP) contents; it used a self‐adaptive partial least squares model (SAM–PLS). The results showed that selected wavenumbers improved the accuracy of prediction for pH, and SOM, TN and AP contents. For the prediction of pH, ATR spectra showed marginally better performance than the other spectra. The DRF spectroscopy was the optimal method for SOM and TN prediction. Available phosphorus contents were predicted poorly by all three spectra. Furthermore, mid‐infrared incident light paths, absorption modes, sensing styles, as well as peak interference and the prediction model, were the main factors that affected the assessment of soil properties in the three types of spectroscopy.Highlights Soil spectra data were measured with DRF, ATR and PAS spectrometers over the range 4000–500 cm−1. A UVE algorithm was used to select effective wavenumbers from the full spectrum. The pH, SOM, TN and AP were estimated using three techniques based on a SAM–PLS model. Mid‐infrared incident light paths and absorption modes affected prediction performance of the three techniques.

Investigating the impact of feature selection on the prediction of solar radiation in different locations in Saudi Arabia

Predictions about the future amount of solar radiation is an important factor that affects the planning and operating of solar energy projects. However, it is a difficult task due to the existence of a high level of uncertainty that is associated with unknown future weather conditions. Accordingly, the prediction process, especially for the short term, requires more attention to unveil hidden relationships and interactions between related variables. Since a large number of parameters affect the estimation and prediction processes, there is a need to apply an effective and efficient feature selection of the input feature space. In this paper, an investigation of different feature selection methods is carried out in order to predict the daily amounts of solar radiation in different locations in Saudi Arabia using a neural networks (NN) predictor. First, the selection of the most important variables is carried out using four different algorithms: ReliefF algorithm, Monte Carlo uninformative variable elimination algorithm (MCUVE), random-frog algorithm, and Laplacian score algorithm (LS). Then, a computational intelligence model of a multi-layer neural network is used as the predictor. The predictor aims to predict the next-day global horizontal irradiance using selected meteorological and solar radiation observations. The experimentation and results of the feature selection methods and the prediction process are described. The results showed the importance of using feature selection methods in order to obtain a reliable prediction of the amount of solar radiation compared with using all the features available.

Maximum relevance, minimum redundancy band selection based on neighborhood rough set for hyperspectral data classification

Band selection is considered to be an important processing step in handling hyperspectral data. In this work, we selected informative bands according to the maximal relevance minimal redundancy (MRMR) criterion based on neighborhood mutual information. Two measures MRMR difference and MRMR quotient were defined and a forward greedy search for band selection was constructed. The performance of the proposed algorithm, along with a comparison with other methods (neighborhood dependency measure based algorithm, genetic algorithm and uninformative variable elimination algorithm), was studied using the classification accuracy of extreme learning machine (ELM) and random forests (RF) classifiers on soybeans’ hyperspectral datasets. The results show that the proposed MRMR algorithm leads to promising improvement in band selection and classification accuracy.

Neighborhood mutual information and its application on hyperspectral band selection for classification

Band selection is considered to be an important processing step in handling hyperspectral data. In this work, we combined Shannon's information entropy with neighborhood rough set and proposed a new measure, called neighborhood mutual information. With the proposed measure which can evaluate the significance of bands for classification, a forward greedy search algorithm for band selection was constructed. To assess the effectiveness of the proposed band selection technique, two classification models (Extreme Learning Machine and Random Forests) were built. The proposed algorithm was compared to neighborhood dependency measure based algorithm, genetic algorithm and uninformative variable elimination algorithm on three (soybean, maize and rice) hyperspectral datasets between 400nm and 1000nm wavelengths. Experimental results show that the proposed method can effectively select key bands and obtain satisfactory classification accuracy.

Uninformative variable elimination for improvement of successive projections algorithm on spectral multivariable selection with different calibration algorithms for the rapid and non-destructive determination of protein content in dried laver

The potential of using partial least square based uninformative variable elimination algorithm (UVEPLS) on successive projections algorithm (SPA) for spectral multivariable selection was evaluated. A case study was done on the visible and shortwave-near infrared (Vis-SNIR) spectroscopy for the rapid and non-destructive determination of protein content in dried laver. Three calibration algorithms, namely multiple linear regression (MLR), partial least square regression (PLS) and least-square support vector machine (LS-SVM), were used for the model establishment based on the selected variables of SPA, UVEPLS and UVEPLS-SPA, respectively. A total of 175 samples were prepared for the calibration (n = 117) and prediction (n = 58) sets. The performances of different pretreatments were compared. Both linear calibration algorithms of MLR and PLS and non-linear calibration algorithms of LS-SVM with linear kernel and RBF kernel obtained similar results based on certain variable selection strategies of SPA, UVEPLS and UVEPLS-SPA. The average improvement percentage of RPD values of four calibration algorithms was 38.66% by calculating SPA on UVEPLS processed variables. Therefore there was much improvement of using UVEPLS on SPA spectral multivariable selection with both linear and nonlinear calibration algorithms in this case. Moreover, the RPD values of both linear and non-linear models based on the thirteen selected variables of UVEPLS-SPA show that coarse quantitative predictions of the protein determination in dried laver is possible based on Vis-SNIR spectra. We hope that the results obtained in this study will help both further chemometric (multivariate selection and calibration analysis) investigations and investigations in the sphere of applied vibrational (Near infrared, Mid-infrared and Raman) spectroscopy of sophisticated multicomponent systems.

Application of a hybrid variable selection method for the classification of rapeseed oils based on 1H NMR spectral analysis

1H nuclear magnetic resonance (NMR) spectroscopy was utilized to distinguish the brands of rapeseed oils. As there are more than four hundreds of NMR variables which can cause the discrimination model redundancy, it is necessary to do effective variable selection. Successive projections algorithm (SPA) executed on the full spectrum only improved a few correct answer rate (CAR) and Cohen’s kappa coefficient (K) compared to full spectrum-least-square support vector machine (LS-SVM) model. The better results of uninformative variable elimination (UVE)-based SPA calculation show that it is necessary to do UVE before SPA. Because the cutoff threshold selection in UVE algorithm using an artificial random noise cannot obtain the optimal results, we applied simulated annealing (SA) algorithm to estimate the optimal cutoff threshold. The discrimination results show that UVE-SA did better works than conventional UVE. Only 13 variables were obtained by UVE-SA-SPA while the conventional UVE-based SPA selected 77 variables. The best 97.5% CAR and K of 0.967 result of UVE-SPA-LS-SVM model show that it is feasible to distinguish different brands of rapeseed oils using 1H NMR spectra. It shows that a combination of SA, UVE, and SPA is effective method for the classification of rapeseed oils. Final result shows that all acyl chains, linolenyl and linoleyl chains, and triglycerides were most important for the classification.