Journal of Chemometrics

A semi-supervised version of Fisher's linear discriminant analysis is developed so that the both labelled and unlabeled observations are used in the model-fitting procedure. This approach is advantageous when few labeled and many unlabeled observations are available. The semi-supervised linear discriminant analysis method is demonstrated on a number of data sets where it is shown to yield better separation of the groups and improved classification over Fisher's linear discriminant analysis.

In this paper, multivariate data analysis was proposed to analyze the impact of fat and muscle on heroin identification based on the profile of spectra of energy dispersive X-ray diffraction. In the space of principal components, results showed that different pure samples clustered in different areas, whereas the location of mixture samples moved between locations of pure samples. The impact of fat and muscle lies in moving the feature points between pure materials in the space of principal components.

This paper focuses on the application of principal component analysis (PCA) to facilitate the optimization of the derivatization of five estrogenic compounds and diethylstilbestrol. So far, only limited information is available on the application of chemometric analyses in the optimization of derivatization, and they are based mainly on the parametric ANOVA test. For the first time, non-parametric statistical method PCA was applied to the experimental data describing the differences in the chemical modification of the estrogenic compounds during derivatization.

Testing the significance of the kth largest eigenvalues of the data covariance matrix is essential in and for many statistical applications. An empirical approach is proposed to extend the results of Johnstone's theorem to the case of autoscaled data matrices. This opens the way to hypothesis testing using the Tracy-Widom distribution in all cases in which autoscaling of the data is required before the calculation of the data covariance matrix, that is, principal component analysis or testing for chance correlations.