Investigating The Methods To Estimate The Number Of Principal Factors In Complex Chemical Systems

Two dimensional measured data in the form of matrix is more and more popular in analytical chemistry with the development of instruments, such as GC-MS data, HPLC-DAD data or the combination of vectors. The data contains rich chemical information. Meanwhile, there exist interference and redundancy, giving rise to the difficulty in analysing data. The crucial point of analysing data is that eliminating the redundancy and extracting main information, namely the dimensionality reduction of data matrix. The number of primary or significant factors is one of the important results of dimensionality reduction. Determining the number of principal factors is a prerequisite step in analyses of matrix-type experimental data; it is also a difficult problem, which is reflected by the coexistence of many methods. A review and a categorization of those methods would provide a panoramic view, and make it possible to generalize important and practically useful information for this problem.This paper elaborates a new classification method, the mathematically rigorous method. The similarity of this kind is told and this kind of method is divided into two categories from a new perspective. The next is my specific work:1. We told about the existing methods of estimating the primary factors’s number at the beginning. Then we employed the computation results of simulated data to analyse criteria of every method in detail. We categorized current methods into three categories which are tagged empirical, mathematically rigorous, and statistical, and summarized the features of every kind’s methods. The mathematically rigorous methods is an innovative point in this paper.2. This section presents the application of every method. We apply these methods to simulated and experimental data. We explored limitations of minor components, overlapping degree of chromatography and noiselevel, providing a reference to each method’s receptance. For experimental data,we predicted the estimates of methods by significant factor’s expectation. We presented the application results not to conclude which method is better and which is bad, but to combine every fit method in order to calculate primary factors more accurately in actual practice.3. We used Raman spectroscopy to measure the Raman spectrum of strontium chloride-solvent(water or methyl alcohol). Then we combined the measured concentration with spectrum data, producting two dimension data. By the employment of chemometric methods to deduce the primary factors in the experimental matrix, we can draw a conclusion that whether there exists solvation. At the same time, we can test the estimation effects of every method better.