The extraction of relevant wavelengths from a large dataset of Near Infrared Spectroscopy (NIRS) is a significant challenge in vibrational spectroscopy research. Nonetheless, this process allows the improvement in the chemical interpretability by emphasizing the chemical entities related to the chemical parameters of samples. With the complexity in the dataset, it may be possible that irrelevant wavelengths are still included in the multivariate calibration. This yields the computational process to become unnecessary complex and decreases the accuracy and robustness of the model. In multivariate analysis, Partial Least Square Regression (PLSR) is a method commonly used to build a predictive model from NIR spectral data. However, in the PLSR method and common commercial chemometrics software, there is no standard wavelength selection procedure applied to screen the irrelevant wavelengths. In this study, a new robust wavelength selection procedure called the modified VIP-MCUVE (mod-VIP-MCUVE) using Filter-Wrapper method and input scaling strategy is introduced. The proposed method combines the modified Variable Importance in Projection (VIP) and modified Monte Carlo Uninformative Variable Elimination (MCUVE) to calculate the scale matrix of the input variable. The modified VIP uses the orthogonal components of Partial Least Square (PLS) in investigating the informative variable in the model by applying the amount of variation both in and , simultaneously. The modified MCUVE uses a robust reliability coefficient and a robust tolerance interval in the selection procedure. To evaluate the superiority of the proposed method, the classical VIP, MCUVE, and autoscaling procedure in classical PLSR were also included in the evaluation. Using artificial data with Monte Carlo simulation and NIR spectral data of oil palm (Elaeis guineensis Jacq.) fruit mesocarp, the study shows that the proposed method offers advantages to improve model interpretability, to be computationally extensive, and to produce better model accuracy.
Multivariate statistical analysis such as partial least square regression (PLSR) is the common data processing technique used to handle high-dimensional data space on near-infrared (NIR) spectral datasets. The PLSR is useful to tackle the multicollinearity and heteroscedasticity problem that can be commonly found in such data space. With the problem of the nonlinear structure in the original input space, the use of the classical PLSR model might not be appropriate. In addition, the contamination of multiple outliers and high leverage points (HLPs) in the dataset could further damage the model. Generally, HLPs contain both good leverage points (GLPs) and bad leverage points (BLPs); therefore, in this case, removing the BLPs seems relevant since it has a significant impact on the parameter estimates and can slow down the convergence process. On the other hand, the GLPs provide a good efficiency in the model calibration process; thus, they should not be eliminated. In this study, robust alternatives to the existing kernel partial least square (KPLS) regression, which are called the kernel partial robust GM6-estimator (KPRGM6) regression and the kernel partial robust modified GM6-estimator (KPRMGM6) regression are introduced. The nonlinear solution on PLSR was handled through kernel-based learning by nonlinearly projecting the original input data matrix into a high-dimensional feature mapping that corresponded to the reproducing kernel Hilbert spaces (RKHS). To increase the robustness, the improvements on GM6 estimators are presented with the nonlinear PLSR. Based on the investigation using several artificial dataset scenarios from Monte Carlo simulations and two sets from the near-infrared (NIR) spectral dataset, the proposed robust KPRMGM6 is found to be superior to the robust KPRGM6 and non-robust KPLS.
1 November 2014 in Journal of Modern Applied Statistical Methods
Traditional inferential procedures often fail with censored and truncated data, especially when sample sizes are small. In this paper we evaluate the performances of the double and single bootstrap interval estimates by comparing the double percentile (DB-p), double percentile-t (DB-t), single percentile (B-p), and percentile-t (B-t) bootstrap interval estimation methods via a coverage probability study when the data is censored using the log logistic model. We then apply the double bootstrap intervals to real right censored lifetime data on 32 women with breast cancer and failure data on 98 brake pads where all the observations were left truncated.
1 May 2018 in Journal of Modern Applied Statistical Methods
The violation of the assumption of homoscedasticity and the presence of high leverage points (HLPs) are common in the use of regression models. The weighted least squares can provide the solution to heteroscedastic regression model if the heteroscedastic error structures are known. Based on Furno (1996), two robust weighting methods are proposed based on HLP detection measures (robust Mahalanobis distance based on minimum volume ellipsoid and diagnostic robust generalized potential based on index set equality (DRGP(ISE)) on robust heteroscedasticity consistent covariance matrix estimators. Results obtained from a simulation study and real data sets indicated the DRGP(ISE) method is superior.
With the complexity of Near Infrared (NIR) spectral data, the selection of the optimal number of Partial Least Squares (PLS) components in the fitted Partial Least Squares Regression (PLSR) model is very important. Selecting a small number of PLS components leads to under fitting, whereas selecting a large number of PLS components results in over fitting. Several methods exist in the selection procedure, and each yields a different result. However, so far no one has been able to determine the more superior method. In addition, the current methods are susceptible to the presence of outliers and High Leverage Points (HLP) in a dataset. In this study, a new automated fitting process method on PLSR model is introduced. The method is called the Robust Reliable Weighted Average—PLS (RRWA-PLS), and it is less sensitive to the optimum number of PLS components. The RRWA-PLS uses the weighted average strategy from multiple PLSR models generated by the different complexities of the PLS components. The method assigns robust procedures in the weighing schemes as an improvement to the existing Weighted Average—PLS (WA-PLS) method. The weighing schemes in the proposed method are resistant to outliers and HLP and thus, preserve the contribution of the most relevant variables in the fitted model. The evaluation was done by utilizing artificial data with the Monte Carlo simulation and NIR spectral data of oil palm (Elaeis guineensis Jacq.) fruit mesocarp. Based on the results, the method claims to have shown its superiority in the improvement of the weight and variable selection procedures in the WA-PLS. It is also resistant to the influence of outliers and HLP in the dataset. The RRWA-PLS method provides a promising robust solution for the automated fitting process in the PLSR model as unlike the classical PLS, it does not require the selection of an optimal number of PLS components.
21 October 2020 in Pertanika Journal of Science and Technology
High Leverage Points (HLPs) are outlying observations in the X -directions. It is very imperative to detect HLPs because the computed values of various estimates are affected by their presence. It is now evident that Diagnostic Robust Generalized Potential which is based on the Minimum Volume Ellipsoid (DRGP(MVE)) is capable of detecting multiple HLPs. However, it takes very long computational running times. Another diagnostic measure which is based on Index Set Equality denoted as DRGP(ISE) is put forward with the main aim of reducing its running time. Nonetheless, it is computationally not stable and still suffers from masking and swamping effects. Hence, in this paper, we propose another version of diagnostic measure which is based on Reweighted Fast Consistent and High Breakdown (RFCH) estimators. We call this measure Diagnostic Robust Generalized Potential based on √n RFCH and it is denoted by DRGP(RFCH). The results of simulation study and real data indicate that our proposed method outperformed the other two methods in term of having the least computing time, highest percentage of correct detection of HLPs and smallest percentage of swamping and masking effects compared to the DRGP(MVE) and DRGP (ISE).