Variable Selection, Outlier Detection, and Figures of Merit Estimation in a Partial Least-Squares Regression Multivariate Calibration Model. A Case Study for the Determination of Quality Parameters in the Alcohol Industry by Near-Infrared Spectroscopy

Abstract

Practical implementation of multivariate calibration models has been limited in several areas due to the requirement of appropriate development and validation to prove their performance to standardization agencies. Herein, a detailed description of the application of multivariate calibration based on partial least-squares regression models (PLSR) for the determination of soluble solids (BRIX), polarizable sugars (POL), and reducing sugars (RS) in sugar cane juice, based on near infrared spectroscopy (NIR), for the alcohol industries is presented. The development of the models, including variable selection and outlier elimination, and their validation by determination of figures of merit, such as accuracy, precision, sensitivity, analytical sensitivity, prediction intervals, and limits of detection and quantification, are described for a representative data set of 1381 sugar cane samples. Values estimated by PLSR are compared with appropriate reference methods, where the results indicated that the PLSR models can be used in the alcohol industry as an alternative to refractometry and lead clarification before polarization measurements (standard methods for BRIX and POL, respectively). For RS, the results of a titration reference method were compared with the PLSR estimates and also with an estimate based on BRIX and POL values, as actually used in the alcohol industry. The PLSR method presented a better agreement with the titration method. However, the results indicated that the RS estimates from both PLSR and those based on the BRIX and POL values, actually used, should be improved to a safe determination of RS.

Soil Organic Carbon Determination Using NIRS: Evaluation of Dichromate Oxidation and Dry Combustion Analysis as Reference Methods in Multivariate Calibration

“…The figures of merit for first order multivariate calibration were described in earlier papers, 12,19,21,25,27 and are not described in detail here. The results for the figures of merit obtained from PLS models are shown in the Table 1.…”

“…In this work, outliers were identified in the models based on data with extreme leverage, unmodeled residuals in spectral data, and unmodeled residuals in the dependent variable. 19 Methods for outliers identification were applied independently for each model. Thus, different samples, in each case, were identified as outliers, which produced optimized models with different numbers of samples into calibration and validation sets.…”

“…The figures of merit were calculated according to previous research. 12,19,21 To evaluate differences between the models produced with dichromate oxidation or dry combustion analysis as reference method, the results of figures of merit for both models were subjected to a paired t-test.…”

Analysis of seized cocaine samples by using chemometric methods and FTIR spectroscopy

“…This is certainly caused by the large chemical variety of the samples, leading to the detection of a high number of outliers (the snowballing effect). 32 Thus, it was decided to build a PLS-DA model to discriminate samples as of high or low content, adopting a threshold value of 15% m/m of cocaine. This value was chosen because it represents a gap, i.e., a small number of samples showed cocaine contents between 15 and 20%, and because it was considered representative to discriminate concentrated and diluted samples by the Brazilian Federal Police.…”

Implementação de um método robusto para o controle fiscal de umidade em queijo minas artesanal. Abordagem metrológica multivariada

“…A detecção de amostras anômalas pode ser feita utilizando diversos critérios, e neste trabalho, uma metodologia robusta 26,27 foi utilizada, avaliando valores de leverage extremos, resíduos elevados no bloco X (dados espectrais) e no bloco Y (valores de referência). Para a avaliação no bloco X, amostras com s(e i ) > 2,0s(e) foram removidas, onde s(e) é o desvio padrão total da informação espectral não modelada e s(e i ) é o desvio padrão especifico de cada amostra.…”