non-parametric class modeling using multivariate curve resolution methods

Class modeling in chemometrics uses statistical models to classify samples by chemical properties. this is used in pharmaceuticals, authentication, environmental monitoring, and food quality. non-normal and non-uniform data require specialized class-modeling methods. potential function methods (pfm) are one of the most efficient probabilistic non-parametric methods [1]. it is a non-parametric method because no assumptions are made about the probability distribution of the samples for a particular class. it is used for discriminant classification and class modeling [2]. the probability distribution for a category is estimated by combining individual contributions from training samples of that class, unlike parametric methods. multivariate curve resolution (mcr) strategies can identify and understand unknown chemical processes and reactions with unknown intermediate species. recently, mcr-als was compared to other methods for distinguishing constituents of two benchmark and two high-dimensional data sets [3]. this work used the efficient data reduction strategy and the mcr method to classify high-dimensional datasets successfully. the efficiency of mcr-da is comparable to that of tested methods such as pls-da [4]. a new mcr-based non-parametric class-modeling strategy is presented here. mcr methods for non-parametric classification were tested on several complex experimental data. the hplc-cad dataset of olive and non-olive oils was used. coupling hplc to a charged aerosol detector (cad) produced 120 chromatogram profiles for olive and non-olive oils. these profiles included 71 olive oil samples from extra virgin, virgin, refined, and pomace categories and 44 edible vegetable oils: canola, maize, flaxseed, grape seed, hazelnut, peanut, rapeseed, safflower, sesame, soybean, and sunflower. this study included five olive-non-olive oil mixtures. olive oil samples were the target, and non-olive and mixed samples were alien. fatty acids in triacylglycerides (tags) chromatographic profiles differentiated olive and non-olive samples [5]. the results were consistent and reliable compared to non-parametric methods like pfm with pca compression and other data classification methods.the proposed mcr approach sometimes results in complex data as good as or even more satisfactory than pca compression. like the original data set, mcr profiles and subspaces compress data accurately. mcr profiles are compound-specific and chemically meaningful, unlike pca representations, and resolving pure component signal weights is beneficial.