Kennard-Stone Algorithm Research Articles

Diesel cetane number estimation from NIR spectra of hydrocracking total effluent

The work shown in this paper offers a fast and efficient alternative for estimating the cetane number of the diesel obtained from the distillation of the hydrocracking total effluent. In this study, the estimation of this diesel property was achieved through a partial least squares regression (PLSR) model using only the NIR spectrum of the hydrocracking total effluent. For calibrating and validating the PLS model, it was used a database containing the NIR spectra acquired on 98 total effluent samples and the cetane number measured on the 98 diesel fractions recovered from each total effluent sample distillation. The database was divided into the calibration and test data sets using the Kennard-Stone algorithm. The regression model developed exhibited good performance in estimating the studied property with errors of calibration (1.3), cross-validation (2.2), and prediction (2.0), close to the reproducibility of the reference method (±3.6). The alternative method for diesel cetane number estimation discussed in this article evidences its feasibility in optimizing diesel fuel characterization by reducing the necessity of the total effluent distillation. Furthermore, the results also show the potential of the alternative proposed to be applied in predicting other properties of fuels obtained from the hydrocracking process.

Proposing Two Local Modeling Approaches for Discriminating PGI Sunite Lamb from Other Origins Using Stable Isotopes and Machine Learning.

For the protection of Protected Geographical Indication (PGI) Sunite lamb, PGI Sunite lamb samples and lamb samples from two other banners in the Inner Mongolia autonomous region were distinguished by stable isotopes (δ13C, δ15N, δ2H, and δ18O) and two local modeling approaches. In terms of the main characteristics and predictive performance, local modeling was better than global modeling. The accuracies of five local models (LDA, RF, SVM, BPNN, and KNN) obtained by the Adaptive Kennard–Stone algorithm were 91.30%, 95.65%, 91.30%, 100%, and 91.30%, respectively. The accuracies of the five local models obtained by an approach of PCA–Full distance based on DD–SIMCA were 91.30%, 91.30%, 91.30%, 100%, and 95.65%, respectively. The accuracies of the five global models were 91.30%, 91.30%, 91.30%, 100%, and 91.30%, respectively. Stable isotope ratio analysis combined with local modeling can be used as an effective indicator for protecting PGI Sunite lamb.

Feature Wavelength Selection Based on the Combination of Image and Spectrum for Aflatoxin B1 Concentration Classification in Single Maize Kernels

Aflatoxin B1 (AFB1) is a very strong carcinogen, maize kernels are easily infected by this toxin during storage. Rapid and accurate identification of AFB1 is of great significance to ensure food safety. In this study, a novel method for classification of AFB1 in single maize kernels was developed. Four groups of maize kernel samples with different AFB1 concentrations (10, 20, 50, and 100 ppb) were prepared by artificial inoculation of toxin. In addition, one group of maize kernel samples without AFB1 were prepared as control, each group with 70 samples. The visible and short wave near-infrared (Vis-SWNIR) region (500–1000 nm) and long wave near-infrared (LWNIR) region (1000–2000 nm) hyperspectral images of all samples were obtained respectively, and the hyperspectral images in 500–2000 nm range was obtained after spectral pretreatment and fusion. Kennard-Stone algorithm was used to divide the samples into calibration set or prediction set. Competitive adaptive reweighted sampling (CARS) and successive projections algorithm (SPA) were used to roughly select the characteristic wavelengths of the calibration set samples, and 25 and 26 effective wavelengths were obtained respectively. Based on the roughly selected wavelengths, a method of fine selection of the characteristic wavelengths was proposed by using the gray-value difference of image (GDI), and a few number of characteristic wavelengths were further selected. Under the LDA classification model, 10 characteristic wavelengths were selected to test the prediction set and the independent verification samples, and the ideal result were obtained with an accuracy of 94.46% and 91.11%, respectively. This study provides a new approach for AFB1 concentration classification of single maize kernels.

Removal of external influences from on-line vis-NIR spectra for predicting soil organic carbon using machine learning

External factors including moisture content negatively affect the prediction accuracy of soil organic carbon (SOC) using on-line visible and near-infrared (vis-NIR) spectroscopy. This study compared the performances of four algorithms to remove the moisture content effect [direct standardization (DS), piecewise direct standardization (PDS), external parameter orthogonalization (EPO), and orthogonal signal correction (OSC)] against non-corrected (NC) spectral models developed with partial least squares regression (PLSR), support vector machine (SVM), random forest (RF), and M5Rules regression. An on-line soil sensing platform coupled with a vis-NIR spectrophotometer (305–1700 nm) was used to scan twelve agricultural fields in Belgium and France. A total of 372 soil samples collected during the on-line measurement were divided into a calibration (260) and a prediction (112) dataset, using the Kennard-Stone algorithm. The latter set together with identical laboratory-measured 112 dry soil spectra formed a transfer dataset to develop EPO, DS and PDS correction matrices. Results showed that models after EPO, PDS and OSC corrections resulted in improved accuracy [coefficient of determination (R2) = 0.60–0.82, root mean square error (RMSE) = 16.1–5.7 g kg−1)], compared to the NC models (R2 = 0.58–0.73, RMSE = 16.5–6.8 g kg−1), whereas the DS (R2 = −0.10 to 0.26, RMSE = 26.8–21.9 g kg−1) provided deteriorated prediction accuracy. The EPO and OSC models provided better prediction accuracy than that of the PDS corrected models. The OSC-M5Rules (R2 = 0.82, RMSE = 5.7 g kg−1) obtained the highest accuracy followed by EPO-M5Rules (R2 = 0.74, RMSE = 6.7 g kg−1) and NC-M5Rules (R2 = 0.73, RMSE = 6.8 g kg−1), which outperformed all PLSR, RF and SVM models. Therefore, on-line vis-NIR spectra should be corrected with the OSC algorithm before calibrating a machine learning model for accurate prediction of SOC.

Quantification of Carbon Dioxide (CO2), Methane (CH4), and Nitrous Oxide (N2O) Using Near Infrared Spectroscopy and Multivariate Calibration in High Humidity Levels

In this work we developed a promising analytical method combining Fourier transform nearinfrared (FT-NIR) spectroscopic technique and first-order multivariate calibration using partial least-squares (PLS) model to simultaneously quantify the main greenhouse gases (GHG’s): methane (CH4), carbon dioxide (CO2), nitrous oxide (N2O) and water vapor (H2O). The models were built using 70 mixtures with different concentration of these gases, 0.25-32.0 ppm to CH4 and N2O, and 50-1100 ppm to CO2 and different values of relative humidity (52-85%, 20 ºC) in synthetic air. After preparing each of the mixtures, they were analyzed by using FT-NIR and a reference analytical technique based on gas chromatography with mass spectrometric detection (GC-MS). The FT-NIR spectrometer was coupled with a long optical path cell, with 105.6 meters of optical path. In sequence, the spectra of all mixtures and its concentration values for each gas were used to build the multivariate calibration models, using PLS regressions. For this, the mixtures were grouped with Kennard Stone algorithm, 50 samples to calibration set and 20 samples to prediction set. The values of RMSEP (root mean square error of prediction) obtained for each model are 0.66, 28.7 and 0.66 ppm, respectively, for CH4, CO2, and N2O. The limits of quantification (LOQ) for each PLS models are 0.26, 3.6, and 0.99 ppm, respectively, for CH4, CO2, and N2O. The results show the potentiality of application of this system to monitoring emission sources in which the concentration of these gases are relatively high, as urban centers, industrial areas, and landfills.

Serum Raman spectroscopy combined with convolutional neural network for label‐free detection of echinococcosis

AbstractEarly disease screening is of great significance to its prevention and treatment, and serum analysis is one of the important methods for early disease screening. In this paper, a portable spectrometer was used to collect serum Raman spectra of echinococcosis patients and healthy volunteers, and a one‐dimensional convolutional neural network (CNN) was used to classify the two types of spectra. The average spectrum shows that the phenylalanine and carotene content in the serum of patients with echinococcosis is less than that in healthy volunteers, whereas the content of amino acids such as proline and tyrosine is higher. We built a CNN model to identify two types of spectra. First, the Kennard–Stone algorithm was used to divide the two types of spectra into a training set and a test set. Afterwards, the optimal hyperparameters of the model were determined by fivefold cross‐validation on the training set. Finally, the test set was used to evaluate the final effect of the model. The accuracy, sensitivity, and specificity of the test set on the model are 94.90%, 90.57%, and 97.11%, respectively. The receiver operating characteristic curve analysis further confirmed the excellent performance of the CNN in the classification of serum Raman spectra. The above results show that serum Raman spectroscopy combined with a one‐dimensional CNN algorithm has great potential in the early diagnosis of echinococcosis.

Soil Classification Based on Deep Learning Algorithm and Visible Near-Infrared Spectroscopy

Changes in land cover will cause the changes in the climate and environmental characteristics, which has an important influence on the social economy and ecosystem. The main form of land cover is different types of soil. Compared with traditional methods, visible and near-infrared spectroscopy technology can classify different types of soil rapidly, effectively, and nondestructively. Based on the visible near-infrared spectroscopy technology, this paper takes the soil of six different land cover types in Qingdao, China orchards, woodlands, tea plantations, farmlands, bare lands, and grasslands as examples and establishes a convolutional neural network classification model. The classification results of different number of training samples are analyzed and compared with the support vector machine algorithm. Under the condition that Kennard–Stone algorithm divides the calibration set, the classification results of six different soil types and single six soil types by convolutional neural network are better than those by the support vector machine. Under the condition of randomly dividing the calibration set according to the proportion of 1/3 and 1/4, the classification results by convolutional neural network are also better. The aim of this study is to analyze the feasibility of land cover classification with small samples by convolutional neural network and, according to the deep learning algorithm, to explore new methods for rapid, nondestructive, and accurate classification of the land cover.

Nondestructive Testing Model of Tea Polyphenols Based on Hyperspectral Technology Combined with Chemometric Methods

Nondestructive detection of tea’s internal quality is of great significance for the processing and storage of tea. In this study, hyperspectral imaging technology is adopted to quantitatively detect the content of tea polyphenols in Tibetan teas by analyzing the features of the tea spectrum in the wavelength ranging from 420 to 1010 nm. The samples are divided with joint x-y distances (SPXY) and Kennard-Stone (KS) algorithms, while six algorithms are used to preprocess the spectral data. Six other algorithms, Random Forest (RF), Gradient Boosting (GB), Adaptive boost (AdaBoost), Categorical Boosting (CatBoost), LightGBM, and XGBoost, are used to carry out feature extractions. Then based on a stacking combination strategy, a new two-layer combination prediction model is constructed, which is used to compare with the four individual regressor prediction models: RF Regressor (RFR), CatBoost Regressor (CatBoostR), LightGBM Regressor (LightGBMR) and XGBoost Regressor (XGBoostR). The experimental results show that the newly-built Stacking model predicts more accurately than the individual regressor prediction models. The coefficients of determination Rc2 andRp2 for the prediction of Tibetan tea polyphenols are 0.9709 and 0.9625, and the root mean square error RMSEC and RMSEP are 0.2766 and 0.3852 for the new model, respectively, which shows that the content of Tibetan tea polyphenols can be determined with precision.

Near infrared spectroscopy determination of chemical and sensory properties in tomato

Fast and massive characterization of quality attributes in tomatoes is a necessary step toward its improvement; for sensory attributes this process is time-consuming and very expensive, which causes its absence in routine phenotpying. We aimed to assess the feasibility of near infrared (NIR) spectroscopy as a fast and economical tool to predict both the chemical and sensory properties of tomatoes. We built partial least squares models from spectra recorded from tomato puree and juice in 53 genetically diverse varieties grown in two environments. Samples were divided in calibration (210 samples for chemical traits, 45 samples for sensory traits) and validation sets (60 and 10, respectively) using the Kennard Stone algorithm. Models from puree spectra gave validation r2 values higher than 0.97 for fructose, glucose, soluble solids content, and dry matter (relative standard error of prediction, RSEP% ranged 3.5–5.8), while r2 values for sensory properties were lower (ranging 0.702–0.917 for taste-related traits (RSEP%: 9.1–20.0), and 0.009–0.849 for texture related traits (RSEP%: 3.6–72.1)). For sensory traits such as explosiveness, juiciness, sweetness, acidity, taste intensity, aroma intensity, and mealiness, NIR spectroscopy is potentially useful for scanning large collections of samples to identify likely candidates to select for tomato quality.

Characterization and identification of different Chinese fermented vinegars based on their volatile components.

In this study, volatile components of 40 Chinese fermented vinegar samples, made from different raw materials, starters, and processing technologies, were collected from different geographic origins in China (Shanxi, Jiangsu, Sichuan, and Fujian Province) and their volatile components were analyzed by headspace-solid-phase microextraction-gas chromatography-mass spectrometry. Sixty-two aroma compounds have been identified by NIST library combined with retention index, mainly including esters, heterocyclics, acids, aldehydes, and ketones. In addition, multivariate analysis including principal component analysis and partial least squares-discriminant analysis (PLS-DA) were carried out to discriminate vinegars based on their composition of volatile components. For PLS-DA models, analysis of variance (ANOVA) or variable importance in the projection (VIP) value were used to select variables with the highest discriminatory power, and the Kennard-Stone algorithm was used to select the training and testing samples. The PLS-DA models (ANOVA or VIP) all provided a classification accuracy of 100% for the training set, and subsequent application of these models allowed the grouping of unknown samples (testing set) according to their characteristics (raw materials and processing technology). PRACTICAL APPLICATIONS: Traditional Chinese vinegars have a long history but nowadays adulterations of them are becoming a problem in the market. In this study, Chinese fermented vinegars from different varieties were identified based on volatile composition. We found that starter cultures and fermentation process have the greatest influence on the volatile components of vinegars, while the influence of raw material and steaming of raw material are weaker volatile components. Then, partial least squares-discriminant analysis models, we carried out could successfully be applied to predict unknown vinegar samples based on a database of volatile components. This study provided a strategy to detect the identity of different vinegars, which can also be used to monitor the quality and safety of traditional Chinese vinegars.

A hybrid optimization method for sample partitioning in near-infrared analysis

The division of calibration and validation is one of the essential procedures that affect the prediction result of the calibration model in quantitative analysis of near-infrared (NIR) spectroscopy. The conventional methods are Kennard-Stone (KS) and sample set partitioning based on joint x-y distances (SPXY). These algorithms use Euclidean distance to cover as many representative samples as possible. This paper proposes an Adaptive Hybrid Cuckoo-Tabu Search (AHCTS) algorithm for partitioning samples based on optimization. The algorithm combines the characteristics of cuckoo search (CS) and tabu search (TS) and fuses with an adaptive function. For comparison, using fishmeal samples as spectral analysis data, KS, SPXY, and AHCTS algorithms were used to divide the modeling samples to establish partial least squares regression (PLSR) models. The experimental results showed that the model established by the proposed algorithm performs better than KS and SPXY. It reveals that the AHCTS method may be an advantageous alternative for quantitative analysis of NIR spectroscopy.

Characterization and authentication of olive, camellia and other vegetable oils by combination of chromatographic and chemometric techniques: role of fatty acids, tocopherols, sterols and squalene

Fatty acids, tocopherols, sterols and squalene were analyzed by chromatographic-based techniques and were selected as variables to build a variety of classification models for the accurate characterization and authentication of olive, camellia oil and six other vegetable oils (soybean, corn, rapeseed, peanut, palm and sunflower). Different unsupervised and supervised chemometrics techniques, such as principal component analysis (PCA), linear discriminant analysis (LDA) and partial least squares discriminant analysis (PLS-DA), have been applied. In addition, the Kennard–Stone algorithm was used to select the training samples for the construction of supervised models. The discriminating power of different components was compared, and the results suggested that fatty acids are the most powerful in distinguishing vegetable oils, followed by tocopherols and sterols, and squalene contributed to the discrimination between olive and camellia oils despite their apparent similarities. This proposed method was straightforward and can be easily implemented to identify unknown oil samples.

Forensic soil source identification: comparing matching by color, vis-NIR spectroscopy and easily-measured physio-chemical properties

This study evaluates to what degree soil samples associated in characteristic space are also close in geographical space, i.e., the possible location from which an unknown sample was obtained in a forensic investigation. The study compares similarity computed from Munsell colors, RGB colors, and full visible-near infrared (vis-NIR) spectra by the spectral angle mapper with similarity based on six easily-measured physio-chemical properties. The reference area is Anhui Province, China with three scales of datasets: provincial, county, and field. Ten diverse “unknown” samples were selected by the Kennard-Stone algorithm from the field-scale dataset and their matches in characteristic space from the several datasets were found by the different methods. The geographic distances of the matches to the “unknowns” were used to evaluate the source identification ability. When a detailed library with local samples is present, a limited set of physio-chemical properties achieved higher geographic accuracy than the color and spectral methods. However, with a regional library the spectral and color methods are superior. Distances in RGB space reveal finer differences than exact matching in Munsell space, but whole-spectra matching outperforms both, because of the rich information influenced by more soil properties than influencing color. We recommend the use of soil vis-NIR spectra as a priority indicator for forensic soil analysis because of its success in this study and its ability to work non-destructively on small quantities of soil.

Exploring multivariate linear regression methods for the prediction of total phenolic content in standard American lager beers using synchronous fluorescence spectroscopy fused data

This work aims to explore chemometric methods for determining the total phenolic content in Brazilian standard American lager beers using synchronous fluorescence spectroscopy fused data. Two hundred and sixteen (216) beer samples were diluted at a ratio of 1:5 (v/v), sonicated for 15 ​min in a water bath, and buffered using an acetic acid/acetate buffer of pH 4.5. The synchronous fluorescence spectra were obtained from 200 to 600 ​nm with a resolution of 1 ​nm. However, different offsets were tested for synchronous fluorescence (Δλ ​= ​20, 50, and 80). All measurements were carried out in triplicate, and the Folin-Ciocalteu method was used as the reference method for the total phenolic content. All chemometric methods were explored using R software and its graphic interface, RStudio. First, the spectral data with different offsets were normalized and fused. After this, the fused data were smoothed with the Savitzky-Golay algorithm with a window of length 11 and a quadratic polynomial, followed by data mean centering. Robust principal component analysis (ROBPCA) was used to identify anomalous data. The Kennard-Stone algorithm was used to split the dataset into 75% for the training set and 25% for the test set. The prediction models used were partial least squares (PLS), orthogonal partial least squares (O-PLS), and elastic net regressions with cross-validation (CV). Ensemble PLS and ensemble sparse PLS with bootstrap resampling were also used. Other PLS regression methods based on the selection of variables were tested, such as significance multivariate regression (sMC – PLS), uninformative variable elimination (UVE – PLS), and interval PLS (iPLS) global and optimized models with cross-validation. All the regression models were built with the training set. The validation was carried out with the application of the regression models to the test set. Within this context, two multivariate regression were highlighted, the CV elastic net and iPLS optimized models, with the root mean square error of prediction (RMSEP) equaling 1.0 and 1.1 ​mg GAE L−1, respectively.

QSAR and Docking Studies on Some Potential Anti-Cancer Agents to Predict their Effect on M14 Melanoma Cell Line

The global incidence of melanoma cancer is increasing rapidly and its metastatic form causes high mortality rates. The research report illustrates the development of a quantitative structure–activity relationship (QSAR) to predict the activities (pGI50) of some anti-cancer agents on melanoma cancer cell line. Subsequently, most potent compounds that showed better pGI50 activities were selected and screened via Lipinski’s rule of five for drug likeness, ADMET risk parameters and lastly docking simulation studies was performed to elucidate their binding mode. Kennard-Stone algorithm was adopted for the data division, the genetic function algorithm was employed for variable selection and multiple linear regressions method was utilized for model development. The derived QSAR model showed, respectively acceptable [( $$R^{2}$$ (0.805), $$R_{adjusted}^{2}$$ (0.773), Qcv2 (0.754) and $$R_{pred}^{2}$$ (0.703)]. The obtained cRP2 for Y-randomization is 0.649, and applicability-domain (A-D) was assessed via leveraged method. Among the screened compounds via Lipinski’s rule of five for oral bioavailability, ADMET risk filter for drug-like features and Docking simulation studies, compound 18 and 25 were identified as the best ligands having the better interactions energies (− 150.679 kcal mol−1 and − 176.246 kcal mol−1) and showed better interactions with the receptor than vemurafenib (− 147.245 kcal mol−1). Thus, the results of this research would be helpful in identification of lead molecule and optimization of novel drug.

Prediction of Soil Organic Carbon in a New Target Area by Near-Infrared Spectroscopy: Comparison of the Effects of Spiking in Different Scale Soil Spectral Libraries.

Near-infrared (NIR) spectroscopy is widely used to predict soil organic carbon (SOC) because it is rapid and accurate under proper calibration. However, the prediction accuracy of the calibration model may be greatly reduced if the soil characteristics of some new target areas are different from the existing soil spectral library (SSL), which greatly limits the application potential of the technology. We attempted to solve the problem by building a large-scale SSL or using the spiking method. A total of 983 soil samples were collected from Zhejiang Province, and three SSLs were built according to geographic scope, representing the provincial, municipal, and district scales. The partial least squares (PLS) algorithm was applied to establish the calibration models based on the three SSLs, and the models were used to predict the SOC of two target areas in Zhejiang Province. The results show that the prediction accuracy of each model was relatively poor regardless of the scale of the SSL (residual predictive deviation (RPD) < 2.5). Then, the Kennard-Stone (KS) algorithm was applied to select 5 or 10 spiking samples from each target area. According to different SSLs and numbers of spiking samples, different spiked models were established by the PLS. The results show that the predictive ability of each model was improved by the spiking method, and the improvement effect was inversely proportional to the scale of the SSL. The spiked models built by combining the district scale SSL and a few spiking samples achieved good prediction of the SOC of two target areas (RPD = 2.72 and 3.13). Therefore, it is possible to accurately measure the SOC of new target areas by building a small-scale SSL with a few spiking samples.

Nondestructive detection of sunset yellow in cream based on near-infrared spectroscopy and interval random forest.

Based on near-infrared spectrum and interval random forest, a fast quantitative analysis method for the content of sunset yellow content was established. The spectra of 132 cream pigment samples were obtained by FT-NIR spectrometer, and various preprocessing methods such as standard normal variable (SNV), wavelet transform (WT), and SG (Savitzky-Golay) were used to smooth and denoise the original spectrum. In this paper, WT and first-order differentiation were used as pretreatment and the Kennard-Stone algorithm was used to divide the data set. Finally interval partial least squares, partial least squares, interval random forest and random forest were used to construct an optimal quantitative analysis model. The experimental results show that the interval random forest can find the best sub-interval to achieve the prediction ability of the model. The R2 (the coefficient of determination) and RMSEP (root mean square error of the prediction) of the prediction set are 0.8965 and 0.2454, respectively. The research results show that near-infrared spectroscopy combined with interval random forest algorithm is a fast and non-destructive method to detect the content of sunset yellow in cream.

Prediction of moisture content of wood using Modified Random Frog and Vis-NIR hyperspectral imaging

Vis-NIR hyperspectral imaging was adopted to predict the moisture content (MC) of wood. A total of 30 beech blocks were used in the experiments, each block was dried 7 times to generate different cases of moisture contents. The 210 samples were divided into a calibration set (140) and a prediction set (70) using Kennard-stone (KS) algorithm. The spectral data of each sample was collected by the hyperspectral imaging system, and pretreated using standard normal variate (SNV). To solve the problem that the random frog algorithm (RF) needs to reset threshold after obtaining the selection probability of all variables at a time, adaptive reweighted sampling (ARS) and exponentially decreasing function (EDF) are applied to improve RF in this paper. The modified RF (MRF) was compared with RF, successive projections algorithm (SPA) and competitive adaptive reweighted sampling (CARS). Based on the full spectrum data (FS) and four characteristic spectral data, different models were built using Gaussian process regression (GPR). The results show that the MRF not only avoids setting thresholds, but also improves the stability and accuracy of the model. The MRF-GPR model achieved the best predictive performance of wood moisture content, with Rp2, RMSEP of 0.9785, 1.6125%, respectively. So the hyperspectral imaging technique is suitable for predicting the moisture content of wood.

A Comprehensive QSAR Study on Antileishmanial and Antitrypanosomal Cinnamate Ester Analogues.

Parasitic infections like leishmaniasis and trypanosomiasis remain as a worldwide concern to public health. Improvement of the currently available drug discovery pipelines for those diseases is therefore mandatory. We have recently reported on the antileishmanial and antitrypanosomal activity of a set of cinnamate esters where we identified several compounds with interesting activity against L. donovani and T. brucei rhodesiense. For a better understanding of such compounds’ anti-infective activity, analyses of the underlying structure-activity relationships, especially from a quantitative point of view, would be a prerequisite for rational further development of such compounds. Thus, quantitative structure-activity relationships (QSAR) modeling for the mentioned set of compounds and their antileishmanial and antitrypanosomal activity was performed using a genetic algorithm as main variable selection tool and multiple linear regression as statistical analysis. Changes in the composition of the training/test sets were evaluated (two randomly selected and one by Kennard-Stone algorithm). The effect of the size of the models (number of descriptors) was also investigated. The quality of all resulting models was assessed by a variety of validation parameters. The models were ranked by newly introduced scoring functions accounting for the fulfillment of each of the validation criteria evaluated. The test sets were effectively within the applicability domain of the best models, which demonstrated high robustness. Detailed analysis of the molecular descriptors involved in those models revealed strong dependence of activity on the number and type of polar atoms, which affect the hydrophobic/hydrophilic properties causing a prominent influence on the investigated biological activities.

Classification of oil pollutants based on excitation-emission matrix fluorescence spectroscopy and two-dimensional discriminant analysis.

Excitation-Emission Matrix (EEM) Fluorescence spectroscopy and two-dimensional (2D) discriminant analysis were used to classify oil pollutants. Firstly, the EEM fluorescence spectroscopy of the oil samples was collected using the FS920 steady-state fluorescence spectrometer, and EEM was preprocessed by removing scattering and normalization. Secondly, EEM was analyzed and characterized by parallel factor analysis (PARAFAC). Finally, all the collected samples were divided into training and test sets by the Kennard-Stone algorithm. The classification models of the training set samples were established by using 2D-PCA, 2D-LDA, PARAFAC-2DPCA, PARAFAC-2DLDA, PARAFAC-LDA and NPLS-DA algorithms, respectively. These models were then used to classify test set samples. The classification performance of the used models was assessed by accuracy, sensitivity and specificity. The best classification results in the used models were obtained by using 2D-PCA and 2D-LDA with 95% and 95% accuracy, respectively. These results provide an important reference for classification of oil pollutants.