Kennard-Stone Method Research Articles

Near-Infrared Spectroscopy with Supervised Machine Learning as a Screening Tool for Neutropenia.

The use of non-invasive tools in conjunction with artificial intelligence (AI) to detect diseases has the potential to revolutionize healthcare. Near-infrared spectroscopy (NIR) is a technology that can be used to analyze biological samples in a non-invasive manner. This study evaluated the use of NIR spectroscopy in the fingertip to detect neutropenia in solid-tumor oncologic patients. A total of 75 patients were enrolled in the study. Fingertip NIR spectra and complete blood counts were collected from each patient. The NIR spectra were pre-processed using Savitzky-Golay smoothing and outlier detection. The pre-processed data were split into training/validation and test sets using the Kennard-Stone method. A toolbox of supervised machine learning classification algorithms was applied to the training/validation set using a stratified 5-fold cross-validation regimen. The algorithms included linear discriminant analysis (LDA), logistic regression (LR), random forest (RF), multilayer perceptron (MLP), and support vector machines (SVMs). The SVM model performed best in the validation step, with 85% sensitivity, 89% negative predictive value (NPV), and 64% accuracy. The SVM model showed 67% sensitivity, 82% NPV, and 57% accuracy on the test set. These results suggest that NIR spectroscopy in the fingertip, combined with machine learning methods, can be used to detect neutropenia in solid-tumor oncology patients in a non-invasive and timely manner. This approach could help reduce exposure to invasive tests and prevent neutropenic patients from inadvertently undergoing chemotherapy.

Sample selection method using near‐infrared spectral information entropy as similarity criterion for constructing and updating peach firmness and soluble solids content prediction models

AbstractWhen using near‐infrared (NIR) techniques for analysis, model construction and maintenance updates are essential. When model construction is performed in machine learning, the sample set is usually divided into the calibration set and the validation set. The representativeness of the calibration set and the reasonable distribution of the validation set affects the accuracy of the established model. In addition, when maintaining and updating models, selecting the most informative updated sample not only improves the model prediction accuracy but also reduces sample preparation. In this paper, the spectral information entropy (SIE) is proposed to be used as a similarity criterion for dividing the sample set and use this criterion to select updated samples. The Kennard–Stone (KS) and the sample set portioning based on joint x–y distance (SPXY) methods were used for comparison to verify the superiority of the proposed method. The results showed that the model built after dividing the sample set using the SIE method has good prediction effect compared with KS and SPXY method. When predicting soluble solid content (SSC) and hardness, the prediction determination coefficient ( ) was improved by more than 15%, and the root mean square error (RMSE) of prediction was reduced by 50%. In terms of model updating, selecting a small number of updated samples using the SIE method can improve the correlation coefficient ( ) to more than 80%, and updated models' prediction accuracy is higher than that of KS and SPXY method. It is confirmed that the SIE method can make the NIR analysis technique more reliable.

Modeling Textural Properties of Cooked Germinated Brown Rice Using the near-Infrared Spectra of Whole Grain.

If a non-destructive and rapid technique to determine the textural properties of cooked germinated brown rice (GBR) was developed, it would hold immense potential for the enhancement of the quality control process in large-scale commercial rice production. We combined the Fourier transform near-infrared (NIR) spectral data of uncooked whole grain GBR with partial least squares (PLS) regression and an artificial neural network (ANN) for an evaluation of the textural properties of cooked germinated brown rice (GBR); in addition, data separation and spectral pretreatment methods were investigated. The ANN was outperformed in the evaluation of hardness by a back extrusion test of cooked GBR using the smoothing combined with the standard normal variate pretreated NIR spectra of 188 whole grain samples in the range of 4000-12,500 cm-1. The calibration sample set was separated from the prediction set by the Kennard-Stone method. The best ANN model for hardness, toughness, and adhesiveness provided R2, r2, RMSEC, RMSEP, Bias, and RPD values of 1.00, 0.94, 0.10 N, 0.77 N, 0.02 N, and 4.3; 1.00, 0.92, 1.40 Nmm, 9.98 Nmm, 1.6 Nmm, and 3.5; and 0.97, 0.91, 1.35 Nmm, 2.63 Nmm, -0.08 Nmm, and 3.4, respectively. The PLS regression of the 64-sample KDML GBR group and the 64-sample GBR group of various varieties provided the optimized models for the hardness of the former and the toughness of the latter. The hardness model was developed by using 5446.3-7506 and 4242.9-4605.4 cm-1, which included the amylose vibration band at 6834.0 cm-1, while the toughness model was from 6094.3 to 9403.8 cm-1 and included the 6834.0 and 8316.0 cm-1 vibration bands of amylose, which influenced the texture of the cooked rice. The PLS regression models for hardness and toughness had the r2 values of 0.85 and 0.82 and the RPDs of 2.9 and 2.4, respectively. The ANN model for the hardness, toughness, and adhesiveness of cooked GBR could be implemented for practical use in GBR production factories for product formulation and quality assurance and for further updating using more samples and several brands to obtain the robust models.

Near-Infrared Model and Its Robustness as Affected by Fruit Origin for ‘Dangshan’ Pear Soluble Solids Content and pH Measurement

Soluble solid content (SSC) and acidity (pH) are two important factors indicating the fruit quality of pears and can be measured by near-infrared spectroscopy (NIRS). However, the robustness of these measurements as affected by different origins of pears remains largely unknown. In this study, we developed an NIRS method to measure ‘Dangshan’ pear (Pyrus spp.) SSC and pH and evaluated the robustness of this non-destructive detection method by examining the effects of pears from three different origins in 2019 and 2020. First, the Kennard–Stone method was used to divide the calibration set of the 2020 pear samples from different orchards. The partial least squares (PLS) model was used to establish the local origin and hybrid origin models to predict the pears’ SSC and pH. Second, a combination of competitive adaptive reweighted sampling (CARS), successive projections algorithm (SPA), and uninformative variable elimination (UVE) was implemented to construct spectral prediction models based on effective variables for assessing the pears’ SSC and pH from local and hybrid origins. The results showed that the local origin detection model produced large errors in predicting the SSC and pH of pears from different origins, and the model, established based on the pear samples of three origins, performed better than the local origin and other hybrid origin models. Finally, the model could be effectively simplified using 70 and 52 characteristic variables selected by the CARS method. Pear samples harvested from three different orchards in 2019 were used as an independent set to verify the validity of the selected characteristic variables. The results showed that the predicted R2p for the SSC and pH measurements of pears of three different origins were more than 0.9 and 0.85, respectively. This finding indicates that the difference in the origin of pears has an important influence on the quantitative inversion of pear SSC and pH measurements, and the combination of the hybrid origin model constructed based on the characteristic variables can improve the prediction accuracy. These findings provide an important theoretical basis for the development of rapid detection devices for the measurements of pears’ SSC and pH.

Kennard-Stone method outperforms the Random Sampling in the selection of calibration samples in SNPs and NIR data

ABSTRACT: Splitting the whole dataset into training and testing subsets is a crucial part of optimizing models. This study evaluated the influence of the choice of the training subset in the construction of predictive models, as well as on their validation. For this purpose we assessed the Kennard-Stone (KS) and the Random Sampling (RS) methods in near-infrared spectroscopy data (NIR) and marker data SNPs (Single Nucleotide Polymorphisms). It is worth noting that in SNPs data, there is no knowledge of reports in the literature regarding the use of the KS method. For the construction and validation of the models, the partial least squares (PLS) estimation method and the Bayesian Lasso (BLASSO) proved to be more efficient for NIR data and for marker data SNPs, respectively. The evaluation of the predictive capacity of the models obtained after the data partition occurred through the correlation between the predicted and the observed values, and the corresponding square root of the mean squared error of prediction. For both datasets, results indicated that the results from KS and RS methods differ statistically from each other by the F test (P-value < 0.01). The KS method showed to be more efficient than RS in practically all repetitions. Also, KS method has the advantage of being easy and fast to be applied and also to select the same samples, which provides excellent benefits in the following analyses.

Meat species identification accuracy improvement using sample set portioning based on joint x-y distance and laser-induced breakdown spectroscopy.

Laser-induced breakdown spectroscopy (LIBS) was suitable for the identification of meat species due to fast and less sample preparation. However, the problem of low accuracy rate of the recognition model caused by improper selection of training set samples by random split has severely restricted the development of LIBS in meat detection. Sample set portioning based on the joint x-y distance (SPXY) method was applied for dividing the meat spectra into a training set and a test set. Then, the five kinds of meat samples (shrimp, chicken, beef, scallop, and pig liver) were classified by the support vector machine (SVM). With the random split method, Kennard-Stone method, and SPXY method, the recognition accuracies of the SVM model were 90.44%, 91.95%, and 94.35%, respectively. The multidimensional scaling method was used to visualize the results of the sample split for the interpretation of the classification. The results showed that the identification performance of the SPXY method combined with the SVM model was best, and the accuracy rates of shrimp, chicken, beef, scallop, and pig liver were 100.00%, 100.00%, 100.00%, 78.57%, and 92.00%, respectively. Moreover, to verify the broad adaptability of the SPXY method, the linear discriminant analysis model, the K-nearest neighbor model, and the ensemble learning model were applied as the meat species identification model. The results demonstrated that the accuracy rate of the classification model can be improved with the SPXY method. In light of the findings, the proposed sample portioning method can improve the accuracy rate of the recognition model using LIBS.

NIR spectroscopy coupled with chemometric algorithms for the prediction of cadmium content in rice samples

Fast determination of heavy metals is necessary and important to ensure the safety of crops. The potential of near-infrared spectroscopy coupled with chemometric technology for quantitative analysis of cadmium in rice was investigated. A total of 825 rice samples were collected and scanned by NIRS. The Kennard–Stone method was applied to divide the samples into calibration and validation sets. Before modeling, the spectrum was preprocessed using first derivation to reduce the baseline shift. Different chemometric tools such as interval partial least squares, moving window partial least squares, synergy interval partial least squares, and backward interval partial least squares were proposed to extract and optimize spectral interval from full-spectrum data. The performance of the calibration models generated on the basis of different regression algorithms was compared and evaluated. Results showed that the PLS models based on four chemometric algorithms outperformed the full-spectrum PLS model. Among the tools, biPLS performed better with the optimal subinterval selection. The root-mean-square error of prediction and correlation coefficient (R) of the biPLS model were 0.2133 and 0.9020, respectively. In addition, the low root-mean-square error of cross-validation was obtained in biPLS, which was 0.1756. NIRS technology combined with biPLS could be considered as an effective and convenient tool for primary screening and measuring of cadmium content in rice. In comparison with classical methodologies, this new technology was beneficial because of its eco-friendliness, fast analysis, and virtually no sample preparation required.

Partial Least Squares Improved Multivariate Adaptive Regression Splines for Visible and Near-Infrared-Based Soil Organic Matter Estimation Considering Spatial Heterogeneity

Under the influence of complex environmental conditions, the spatial heterogeneity of soil organic matter (SOM) is inevitable, and the relationship between SOM and visible and near-infrared (VNIR) spectra has the potential to be nonlinear. However, conventional VNIR-based methods for soil organic matter estimation cannot simultaneously consider the potential nonlinear relationship between the explanatory variables and predictors and the spatial heterogeneity of the relationship. Thus, the regional application of existing VNIR spectra-based SOM estimation methods is limited. This study combines the proposed partial least squares–based multivariate adaptive regression spline (PLS–MARS) method and a regional multi-variable associate rule mining and Rank–Kennard-Stone method (MVARC-R-KS) to construct a nonlinear prediction model to realize local optimality considering spatial heterogeneity. First, the MVARC-R-KS method is utilized to select representative samples and alleviate the sample global underrepresentation caused by spatial heterogeneity. Second, the PLS–MARS method is proposed to construct a nonlinear VNIR spectra-based estimation model with local optimization based on selected representative samples. PLS–MARS combined with the MVARC-R-KS method is illustrated and validated through a case study of Jianghan Plain in Hubei Province, China. Results showed that the proposed method far outweighs some available methods in terms of accuracy and robustness, suggesting the reliability of the proposed prediction model.

An rapid nondestructive testing method for distinguishing rice producing areas based on Raman spectroscopy and support vector machine

This study reports method for rapid and nondestructive identification of single-grain rice produced in neighboring areas. Four types of representative rice from different areas were selected as experimental samples. A total of 284 single-grain rice Raman spectra were acquired and spectral information in the 400–1700 cm−1 spectral area was extracted for analysis. First, the samples were divided into 190 calibration sets and 94 validation sets by the Kennard–Stone method, and the raw spectra were pretreated using smoothing and differential methods. Next, principal component analysis (PCA) and a successive projections algorithm (SPA) were used to extract the optimal principal components and effective wavelengths, which were used as the input variables for k-NearestNeighbor (KNN) and least-squares support vector machine (LS-SVM) algorithms. The PCA-KNN,SPA-KNN,PCA-LS-SVM and SPA-LS-SVM models were established based on information from the rice spectra. Finally, the model was applied to classify the validation set samples. The recognition accuracies of the PCA-KNN,SPA-KNN, PCA-LS-SVM and SPA-LS-SVM models for the validation set are 91.43 %,93.62 %,91.49 and 94.68 %, respectively. These results indicate that single-grain rice from different producing areas can be identified by using Raman spectroscopy combined with KNN and LS-SVM. This method can achieve rapid and completely non-destructive testing.

Selection of a calibration sample subset by a semi-supervised method

For spectroscopic measurements, representative samples are needed in the course of building a calibration model to guarantee accurate predictions. The most widely used selection method is the Kennard-Stone method, which can be used before a reference measurement is done. In this paper, a method termed semi-supervised selection is presented to determine whether a sample should be added to the calibration set. The selection procedure has two steps. First, part of the population of samples is selected using the Kennard-Stone method, and their concentrations are measured. Second, another part of the population of samples is selected based on the scalar value distribution of the net analyte signal. If the net analyte signal of a sample is distinctive compared to the existing net analyte signal values, then the sample is added to the calibration set. The analyte of interest in the sample is then measured so that the sample can be used as a calibration sample. By a validation test, it is shown that the presented method is more efficient than random selection and Kennard-Stone selection. As a result, both the time and the money spent on reference measurements are saved.

Evaluation of spectral pretreatments, spectral range, and regression methods for quantitative spectroscopic analysis of soil organic carbon composition

ABSTRACTAlthough there is an increasing interest in using infrared spectroscopy for the simple, rapid, and inexpensive prediction of soil organic carbon content, few studies have used this technique to measure organic carbon chemistry. In this paper, based on both near-infrared and mid-infrared diffuse reflectance spectroscopy, we compared the use of instrumentation, spectral pretreatment, and regression method for the prediction of three parameters related to organic carbon content, one related to isotopic composition, and five related to organic carbon chemistry. A total of 140 soil samples collected from seven oriental oak forest sites across East China were used as the data set for the calibration-validation procedure. Calibrations using sample set partitioning based on joint x-y distances method significantly outperformed those using Kennard-Stone method. Compared to models using linear method (i.e., partial least squares), those using non-linear regression method (i.e., support vector machines) greatly improved the prediction precision of the alkyl-to-O-alkyl ratio and performed slightly better for the other organic carbon chemical compositions. Instrumentation had a large effect as mid-infrared models had higher average prediction accuracies than near-infrared models. We finally proposed a model using second derivative preprocessing, joint x-y distances based sample set partitioning, mid-infrared spectra, and support vector machines regression to quantify organic carbon chemistry in this study. The results are helpful for the further study of soil composition measurement.

Identifying CPPU-Treated Kiwifruits Using Near-Infrared Hyperspectral Imaging Technology

The application of forchlorfenuron (N-(2-chloro-4-pyridyl)-N′-phenylurea), short form CPPU, on kiwifruits has become an important factor that influences kiwifruit economic efficiency and the health development of kiwifruit industry. This study aims to investigate the feasibility of using hyperspectral imaging technology to identify kiwifruits treated with CPPU (named as treated kiwifruits) from kiwifruits without CPPU treatment (named as untreated kiwifruits), and to investigate which model, developed for a single variety or for two varieties together, has better identification performance. Two hundred and forty “Xixuan” kiwifruits and 240 “Huayou” kiwifruits (120 treated kiwifruits and 120 untreated kiwifruits for each variety) were used to obtain hyperspectral from 865.11 to 1711.71 nm. The samples were divided into calibration set and prediction set based on Kennard–Stone method as the ratio of 3:1. Standard normal variate transformation was used to preprocess obtained spectra. Successive projections algorithm (SPA) was applied to extract the characteristic wavelengths from full spectra (FS). Support vector machine (SVM) and extreme learning machine (ELM) modeling methods were used to establish identification models of treated kiwifruits based on FS, characteristic wavelengths extracted by SPA, and universal wavelengths (UWs) extracted from characteristic wavelengths selected by SPA for “Xixuan,” “Huayou,” and two varieties together, respectively. The results showed that the number of characteristic wavelengths selected by SPA were 18, 18, and 21 for “Xixuan,” “Huayou,” and two varieties together, respectively. Five UWs were found for the three different samples. The best model was SPA-ELM for both “Xixuan” (99.8 % accuracy rate for predication set) and “Huayou” (100.0 % accuracy rate for predication set). Models of SPA-SVM and SPA-ELM, whose accuracy rate reached 100 % for both calibration and predication sets, had the best performance when the two varieties were used together. The performances of models built for two varieties together were better than that for “Xixuan,” but they were worse than that for “Huayou.”.The study indicates that NIR hyperspectral imaging technique can be used as a noninvasive method for identifying CPPU-treated kiwifruits from untreated ones, and it is potential to develop a model based on multi-varieties together.

Optimization of Informative Spectral Regions in FT-NIR Spectroscopy for Measuring the Soluble Solids Content of Apple

A novel potential method, linear combination weight PLS (LCW-PLS) model, was suggested for improving the performance of routine PLS model based on selected informative regions. Moving window partial least squares (MWPLS), genetic algorithms interval partial least squares (GAiPLS) and synergy interval partial least squares (SiPLS) were used to optimize informative spectral regions from FT-NIR spectra. A total of 660 apples harvested at 2006, 2007 and 2008, were divided into calibration and prediction sets by Kennard-Stone method. The best calibration model was obtained by LCW-PLS method based on informative spectral regions of 4328–4787, 5323–5512, 5982–7135 and 7159–7463 cm− 1 selected by MWPLS procedure, and corresponding weights of 0.004, 0.070, 0.066 and 0.860, respectively. The LCW-MWPLS model was applied to predict samples, the prediction results were with RP of 0.942, RMSEP of 0.649 %Brix and RPDP of 3.10. In addition, developed LCW-MWPLS model using random two years samples was used to predict one ...

Identification of Kiwifruits Treated with Exogenous Plant Growth Regulator Using Near-Infrared Hyperspectral Reflectance Imaging

The goal of this study was to explore the potential of using near-infrared (NIR) hyperspectral imaging in combination with multivariate analysis for the identification of kiwifruits treated with exogenous plant growth regulator (EPGR). One hundred twenty kiwifruits, variety ‘Xixuan,’ treated with EPGR and another 120 ‘Xixuan’ kiwifruits without EPGR (named as normal ones) were used in the study. Hyperspectral images were acquired using a near-infrared hyperspectral imaging system in the spectral range of 865.11–1,711.71 nm. Based on the Kennard-Stone method, the samples were divided into two sets: 180 samples in calibration set and 60 samples in validation set. Standard normal variate transformation was used to preprocess obtained spectra. Principal component analysis (PCA) and successive projections algorithm (SPA) were used to select principal components and characteristic wavelengths. Partial least square (PLS) regression and support vector machine (SVM) modeling methods were used to establish models for identifying EPGR-treated kiwifruits and normal ones. The results indicated that average correct identification rates of all models were higher than 98.9 and 96.7 % for the calibration set and validation set, respectively. The identification performance of PLS was better than that of SVM, and the best model was PLS-SPA, whose average accuracy rate reached 100 % for the calibration set and 98.4 % for the validation set. The results demonstrated that NIR hyperspectral imaging technique can be used as a noninvasive method for distinguishing kiwifruits treated with EPGR from normal ones.

Protein determination in serum and whole blood by attenuated total reflectance infrared spectroscopy

Attenuated total reflectance mid-infrared spectra of serum and blood samples were obtained from 4,000 to 600 cm(-1). Models for the determination of albumin, immunoglobulin, total globulin, and albumin/globulin coefficients were established for serum samples, using reference data obtained by capillary electrophoresis. Based on the use of the amide bands I and II regions, the relative root mean square error of prediction (RRMSEP) was 4.9, 14.9, 4.5, and 7.1% for albumin, immunoglobulin, total globulin, and albumin/globulin coefficients, respectively, determined in an independent validation set of 120 samples using 200 samples for calibration. Additionally, the use of Kennard-Stone method for the selection of a representative calibration subset of samples provided comparable results using only 60 samples. For whole blood analysis, hemoglobin was determined in 40 validation samples using models built from 40 calibration independent samples with RRMSEP of 8.3, 5.5, and 4.9% with models built from direct spectra in the first case and from sample spectra recorded after lysis by sodium dodecyl sulfate and freezing, respectively, for the last two ones. The developed methodologies offer green alternatives for patient diagnosis in a few minutes, minimizing the use of reagents and residues and being adaptable for its use as a point-of-care method.

Prediction of wood property in Chinese Fir based on visible/near-infrared spectroscopy and least square-support vector machine

A method for the quantification of density of Chinese Fir samples based on visible/near-infrared (vis–NIR) spectrometry and least squares-support vector machine (LS-SVM) was proposed. Sample set partitioning based on joint x– y distances (SPXY) algorithm was used for dividing calibration and prediction samples, it is of value for prediction of property involving complex matrices. A stepwise procedure is employed to select samples according to their differences in both x (instrumental responses) and y (predicted parameter) spaces. For comparison, the models were also constructed by Kennard–Stone method, as well as by using the duplex and random sampling methods for subset partitioning. The results revealed that the SPXY algorithm may be an advantageous alternative to the other three strategies. To validate the reliability of LS-SVM, comparisons were made among other modeling methods such as support vector machine (SVM) and partial least squares (PLS) regression. Satisfactory models were built using LS-SVM, with lower prediction errors and superior performance in relation to SVM and PLS. These results showed possibility of building robust models to quantify the density of Chinese Fir using near-infrared spectroscopy and LS-SVM combined SPXY algorithm as a nonlinear multivariate calibration procedure.