Iterative predictor weighting (IPW) PLS: a technique for the elimination of useless predictors in regression problems

Forina, Michele; Casolino, Chiara; Millán, Carlos

doi:10.1002/(sici)1099-128x(199903/04)13:2<165::aid-cem535>3.0.co;2-y

Cited by 134 publications

(48 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…However, this stochastic nature (involving predictable processes as well as random actions) is also a major disadvantage in establishing a universal spectral subset since it is almost impossible to recreate identical GA or SA models. 26 Other methods that have been employed for feature selection include iterative variable selection, 27 iterative predictor weighting, 28 and uninformative variable elimination. 29 In particular, the important work of Centner et al 29 proposes the addition of artificial (noise) variables prior to the development of a closed form PLS or principal component regression (PCR) model for the dataset containing both the experimental and artificial variables.…”

Section: Theoretical Background 21 Wavelength Selectionmentioning

confidence: 99%

Wavelength selection-based nonlinear calibration for transcutaneous blood glucose sensing using Raman spectroscopy

et al. 2011

View full text Add to dashboard Cite

Abstract. While Raman spectroscopy provides a powerful tool for noninvasive and real time diagnostics of biological samples, its translation to the clinical setting has been impeded by the lack of robustness of spectroscopic calibration models and the size and cumbersome nature of conventional laboratory Raman systems. Linear multivariate calibration models employing full spectrum analysis are often misled by spurious correlations, such as system drift and covariations among constituents. In addition, such calibration schemes are prone to overfitting, especially in the presence of external interferences that may create nonlinearities in the spectra-concentration relationship. To address both of these issues we incorporate residue error plot-based wavelength selection and nonlinear support vector regression (SVR). Wavelength selection is used to eliminate uninformative regions of the spectrum, while SVR is used to model the curved effects such as those created by tissue turbidity and temperature fluctuations. Using glucose detection in tissue phantoms as a representative example, we show that even a substantial reduction in the number of wavelengths analyzed using SVR lead to calibration models of equivalent prediction accuracy as linear full spectrum analysis. Further, with clinical datasets obtained from human subject studies, we also demonstrate the prospective applicability of the selected wavelength subsets without sacrificing prediction accuracy, which has extensive implications for calibration maintenance and transfer. Additionally, such wavelength selection could substantially reduce the collection time of serial Raman acquisition systems. Given the reduced footprint of serial Raman systems in relation to conventional dispersive Raman spectrometers, we anticipate that the incorporation of wavelength selection in such hardware designs will enhance the possibility of miniaturized clinical systems for disease diagnosis in the near future. C 2011 Society of Photo-Optical Instrumentation Engineers (SPIE).

show abstract

Section: Theoretical Background 21 Wavelength Selectionmentioning

confidence: 99%

Wavelength selection-based nonlinear calibration for transcutaneous blood glucose sensing using Raman spectroscopy

et al. 2011

View full text Add to dashboard Cite

show abstract

“…13 Weights of samples are dependent upon the value of the cross-validated regression residuals. The median r˜ of the absolute values of the residuals |ri| is calculated and the new sample square weights can be obtained by the weight function:…”

Section: Modified Ipow-pls Methodsmentioning

confidence: 99%

“…Weights of samples are dependent upon the values of the regression residuals, as done in iterative reweighted PLS (IRPLS), 12 and weights of wavelengths are obtained from the PLS regression coefficients and the standard deviation, as done in iterative predictor weighting PLS (IPW-PLS). 13 During the calculation of the weights of samples, the tuning constant in IPOW-PLS is the key parameter that defines a threshold beyond which a weight of zero is assigned to that sample. A different tuning constant for the same weight function can give different result, 7,8 and the data-dependent feature makes IPOW-PLS inconstant and computationally complex.…”

Section: Introductionmentioning

confidence: 99%

Simultaneous Wavelength Selection and Outlier Detection in Multivariate Regression of Near-Infrared Spectra

Chen

Shao

et al. 2005

ANAL. SCI.

View full text Add to dashboard Cite

IntroductionNear-infrared (NIR) spectroscopy is a powerful analytical tool widely used to measure, both directly and indirectly, a large number of chemical and physical properties.1,2 However, the presence of the relatively weak and highly overlapping spectral bands in the NIR spectra requires the use of multivariate calibration methods to extract the analytical information accurately. To construct a multivariate calibration model, one needs to measure the NIR spectra and the corresponding properties of samples in order to capture the variation in the sample properties to be predicted. 3 Frequently, the number of wavelength variables is much larger than that of concentration variables in order to get as more information as possible. 4 Inevitably, many wavelengths in the NIR spectra are useless, and the presence of them will degrade the performance of the regression model. 5 On the other hand, in practical situations, unexpected experimental or measurement noise is inevitably introduced into concentration (or dependent) variables as well as spectral data, 6 so it is reasonable to consider that the outliers introduced in the calibration set will be more prevalent. Outliers are data points that have a rather large influence on the regression solution and the occurrence of such data points can lead to considerable deviations from normality. 8 In this regard, many approaches for selection of wavelengths or detection of outliers have been proposed. Most authors focus on these problems separately. Up to now, very few methods have been adopted for eliminating both the uninformative wavelengths and the outliers because of some complexity. [9][10][11] Among few methods available, only the method called "iterative predictors and objects weighting PLS" (IPOW-PLS) 9 is applicable to NIR spectra. The method can assign suitable weights to both wavelengths and samples in the iterations of PLS1 algorithm. Weights of samples are dependent upon the values of the regression residuals, as done in iterative reweighted PLS (IRPLS), 12 and weights of wavelengths are obtained from the PLS regression coefficients and the standard deviation, as done in iterative predictor weighting PLS (IPW-PLS). 13 During the calculation of the weights of samples, the tuning constant in IPOW-PLS is the key parameter that defines a threshold beyond which a weight of zero is assigned to that sample. A different tuning constant for the same weight function can give different result, 7,8 and the data-dependent feature makes IPOW-PLS inconstant and computationally complex. To solve this problem, we propose a new criterion in the modified version of IPOW-PLS to determine objectively the level at which the sample can be removed as an outlier or not. The criterion is based on the commonly used "3σ edit rule". 6 Another disadvantage of IPOW-PLS is that the calculation of IPOW-PLS will be rather time-consuming when the number of wavelengths is large, because IPOW-PLS is based on the iterations of PLS algorithm. In order to expedite the IPOW-PLS calculation, we pro...

show abstract

“…In some cases, conventional methods may not offer a satisfactory solution to a given problem due to the complexity of the data, and it may be necessary to apply some sort of variable selection. There have been many mathematical methods of variable selection and genetic algorithm is one of them offering a fast and effective solution for large scale problems [17][18][19][20]. Inverse least squares (ILS) is based on the inverse of Beer's Law, where concentrations of an analyte are modeled as a function of absorbance measurements.…”

Section: Introductionmentioning

confidence: 99%

Determination of aluminum rolling oil additives and contaminants using infrared spectroscopy coupled with genetic algorithm based multivariate calibration

Yalcin

Ergun²,

Uçar³

et al. 2010

Vibrational Spectroscopy

View full text Add to dashboard Cite

a b s t r a c tGenetic algorithm based multivariate calibration models were generated for infrared spectroscopic determination of aluminum rolling oil additives and contaminants such as gear and hydraulic oils. Two different additives and six different suspected contaminants were investigated in the base oil lubricant. Routine analysis samples from 9 different aluminum rolling systems were collected in a period of 2 months in an aluminum rolling plant and gas chromatography (GC) is used as the reference method. Infrared absorbance spectra of the samples were then collected and the reference values obtained with GC were used together with these spectra for model building. Inverse least squares method was optimized with a genetic algorithm by selecting the most contributing regions of the infrared spectra for each component. The R 2 values between GC and multivariate spectroscopic determinations were around 0.99 indicating a good correlation between the two methods. Performance of genetic algorithm based multivariate calibration models were also compared with partial least squares (PLS) method. The study showed that infrared spectroscopy coupled with multivariate calibration can be used for continuous monitoring of additives and contaminants in aluminum rolling oil. By this way, analysis time is significantly reduced and simultaneous determination of all the components can be accomplished.

show abstract

Iterative predictor weighting (IPW) PLS: a technique for the elimination of useless predictors in regression problems

Cited by 134 publications

References 12 publications

Wavelength selection-based nonlinear calibration for transcutaneous blood glucose sensing using Raman spectroscopy

Wavelength selection-based nonlinear calibration for transcutaneous blood glucose sensing using Raman spectroscopy

Simultaneous Wavelength Selection and Outlier Detection in Multivariate Regression of Near-Infrared Spectra

Determination of aluminum rolling oil additives and contaminants using infrared spectroscopy coupled with genetic algorithm based multivariate calibration

Contact Info

Product

Resources

About