• Journal of the Korean Data and Information Science Society
• /
• 제14권2호
• /
• pp.355-366
• /
• 2003

A unified procedure for principal component regression (PCR), partial least squares (PLS) and ordinary least squares (OLS) is proposed. The process gives solutions for PCR, PLS and OLS in a unified and non-iterative way. This enables us to see the interrelationships among the three regression coefficient vectors, and it is seen that the so-called E-matrix in the solution expression plays the key role in differentiating the methods. In addition to setting out the procedure, the paper also supplies a robust numerical algorithm for its implementation, which is used to show how the procedure performs on a real world data set.

• Journal of the Korean Data and Information Science Society
• /
• 제18권2호
• /
• pp.327-344
• /
• 2007

다중공선성의 데이터에 사용되는 대표적인 편향회귀방법은 능형회귀(RR), 주성분회귀(PCR), 부분최소제곱회귀(PLS) 등이다. 이 회귀방법들은 계수베거 추정량의 놈(norm)이 모두 보통 최소제곱회귀(OLS)의 추정량의 놈보다 작아진다는 의미에서 축소회귀라 부른다. 새로운 회귀방법으로 RR과 PCR을 결합한 능형주성분회귀(RPCR)가 있고 RR과 PLS를 결합한 능형부분최소제곱회귀(RPLS)가 있으며 이들도 또한 축소회귀이다. 이들 추정량은 X'X의 고유벡터들의 선형결합으로 나타낼 수 있고 따라서 각 고유방향에서 OLS에 비해 얼마나 축소되는지를 연구할 수 있다. 본 논문에서는 먼저 이들 추정량을 일반적인 축소인자의 식으로 나타내고 이를 이용하여 MSE의 일반식을 구하였으며 PLS 추정량의 MSE 식도 구하였다. 그리고 RPLS의 축소인자 식을 두 가지 다른 형태로 유도하였다. RPLS의 경우도 이 축소인자 식을 MSE의 일반식에 대입하면 MSE 식이 바로 얻어진다. 그러나 PLS나 RPLS의 축소인자는 y의 복잡한 비선형이 되어 결정적이 아니므로 이들 추정량의 MSE는 근사적인 식이라 할 수 있다. 따라서 PLS나 RPLS를 평가하기 위해 이 MSE를 사용하는 것은 제한적이며, 경험적인 방법으로 이들 회귀의 수행성을 평가하는 것이 필요하다. 다중공선성의 대표적인 데이터인 근적외선 분광 데이터를 이용하여 이 유도된 회귀의 축소인자 값이 인자수에 따라 어떻게 변화하는지와 전체적인 축소 비율도 살펴보았다. 이들의 축소 형태를 잘 이해하면 회귀방법들의 예측력과 안정성을 파악하는데 많은 도움이 되리라 판단된다.

• Journal of the Korean Data and Information Science Society
• /
• 제18권1호
• /
• pp.51-61
• /
• 2007

Ridge regression (RR), principal component regression (PCR) and partial least squares regression (PLS) are among popular regression methods for collinear data. While RR adds a small quantity called ridge constant to the diagonal of X'X to stabilize the matrix inversion and regression coefficients, PCR and PLS use latent variables derived from original variables to circumvent the collinearity problem. One problem of PCR and PLS is that they are very sensitive to overfitting. A new regression method is presented by combining RR and PCR and PLS, respectively, in a unified manner. It is intended to provide better predictive ability and improved stability for regression models. A real-world data from NIR spectroscopy is used to investigate the performance of the newly developed regression method.

• 산업공학
• /
• 제23권3호
• /
• pp.229-238
• /
• 2010

Virtual metrology (VM), a critical component of semiconductor manufacturing, is an efficient way of assessing the quality of wafers not actually measured. This is done based on a model between equipment sensor data (obtained for all wafers) and the quality characteristics of wafers actually measured. This paper considers principal component regression (PCR), partial least squares regression (PLSR), kernel PCR (KPCR), and kernel PLSR (KPLSR) as VM models. For each regression model, two cases are considered. One utilizes all explanatory variables in developing a model, and the other selects significant variables using the genetic algorithm (GA). The prediction performances of 8 regression models are compared for the short- and long-term etch process data. It is found among others that the GA-KPLSR model performs best for both types of data. Especially, its prediction ability is within the requirement for the short-term data implying that it can be used to implement VM for real etch processes.

• Journal of the Korean Data and Information Science Society
• /
• 제15권2호
• /
• pp.307-316
• /
• 2004

Principal Component Regression(PCR) and Partial Least Squares Regression(PLSR) are the two most popular regression techniques in chemometrics. In the field of chemometrics usually the number of regressor variables greatly exceeds the number of observation. So we have to reduce the number of regressors to avoid the identifiability problem. In this paper we compare PCR and PLSR techniques combined with various robust regression methods including regression depth estimation. We compare the efficiency, goodness-of-fit and robustness of each estimators under several contamination schemes.

• 대한인간공학회지
• /
• 제20권1호
• /
• pp.45-62
• /
• 2001

This study suggests a genetic algorithm-based partial least squares (GA-based PLS) method to select the design variables for building a usability model. The GA-based PLS uses a genetic algorithm to minimize the root-mean-squared error of a partial least square regression model. A multiple linear regression method is applied to build a usability model that contains the variables seleded by the GA-based PLS. The performance of the usability model turned out to be generally better than that of the previous usability models using other variable selection methods such as expert rating, principal component analysis, cluster analysis, and partial least squares. Furthermore, the model performance was drastically improved by supplementing the category type variables selected by the GA-based PLS in the usability model. It is recommended that the GA-based PLS be applied to the variable selection for developing a usability model.

• 제어로봇시스템학회논문지
• /
• 제9권2호
• /
• pp.159-167
• /
• 2003

This study suggests the hybrid fault diagnosis method of signed digraph (SDG) and partial least squares (PLS). SDG offers a simple and graphical representation for the causal relationships between process variables. The proposed method is based on SDG to utilize the advantage that the model building needs less information than other methods and can be performed automatically. PLS model is built on local cause-effect relationships of each variable in SDG. In addition to the current values of cause variables, the past values of cause and effect variables are inputted to PLS model to represent the Process armies. The measured value and predicted one by dynamic PLS are compared to diagnose the fault. The diagnosis example of CSTR shows the proposed method improves diagnosis resolution and facilitates diagnosis of masked multiple-fault.

• 한국정보과학회:학술대회논문집
• /
• 한국정보과학회 2001년도 가을 학술발표논문집 Vol.28 No.2 (1)
• /
• pp.598-600
• /
• 2001

DNA chip technology enables us to monitor thousands of gene expressions per sample simultaneously. Typically, DNA microarray data has at least several thousands of variables (genes) wish relatively smal1 number of samples. Thus feature (gene) selection by dimensionality reduction is necessary for efficient data analysis. In this paper we employ the partial least squares (PLS) method for gene selection and the principal component analysis (PCA) method for classification. The useful behavior of the PLS is verified by computer simulations.

• 제어로봇시스템학회논문지
• /
• 제10권2호
• /
• pp.185-191
• /
• 2004

Partial least squares(PLS) is one of multiplicate statistical process methods and has been developed in various algorithms with the characteristics of principal component analysis, dimensionality reduction, and analysis of the relationship between input variables and output variables. But it has been limited somewhat by their dependency on linear mathematics. The algorithm is proposed to classify for the non-linear data using PLS and the residual compensator(RC) based on radial basis function network (RBFN). It compensates for the error of the non-linear data using the RC based on RBFN. The experimental result is given to verify its efficiency compared with those of previous works.

• Bulletin of the Korean Chemical Society
• /
• 제25권5호
• /
• pp.647-651
• /
• 2004

Partial Least Squares (PLS) calibration performance has been systematically investigated by changing spectral resolutions of near-infrared (NIR) spectra. For this purpose, synthetic samples simulating naphtha were prepared to examine the calibration performance in complex chemical matrix. These samples were composed of $C_6-C_9$ normal paraffin, iso-paraffin, naphthene, and aromatic hydrocarbons. NIR spectra with four different resolutions of 4, 8, 16, and 32$cm^{-1}$ were collected and then PLS regression was performed. For PLS calibration, five different group compositions (such as total paraffin content) and six different pure components (such as benzene concentration) were selected. The overall results showed that at least 8$cm^{-1}$ resolution was required to resolve the complex chemical matrix such as naphtha. It was found that the influence of resolution on the PLS calibration was varied by the spectral features of a component.