Class-modelling methods

A first distinction which is often made is that between methods focusing on discrimination and those that are directed towards modelling classes. Most methods explicitly or implicitly try to find a boundary between classes. Some methods such as linear discriminant analysis (LDA, Sections 33.2.2 and 33.2.3) are designed to find explicit boundaries between classes while the k-nearest neighbours (A -NN, Section 33.2.4) method does this implicitly. Methods such as SIMCA (Section 33.2.7) put the emphasis more on similarity within a class than on discrimination between classes. Such methods are sometimes called disjoint class modelling methods. While the discrimination oriented methods build models based on all the classes concerned in the discrimination, the disjoint class modelling methods model each class separately. 

Up to this point the methods of classification operate in the same way. They differ considerably, however, in the way that rules for classification are derived. In this regard the various methods are of three types 1) class discrimination or hyperplane methods, 2) distance methods, and 3) class modeling methods. 

Only one class modeling method is conmonly applied to analytical data and this is the SIMCA method ( ) of pattern recognition. In this method the class structure (cluster) is approximated by a point, line, plane, or hyperplane. Distances around these geometric functions can be used to define volumes where the classes are located in variable space, and these volumes are the basis for the classification of unknowns. This method allows the development of information beyond class assignment ( ). 

In SIMCA, a class modeling method, a parameter called modeling power is used as the basis of feature selection. This variable is defined in Equation 4, where is the standard deviation of a vari-... 

Since SIMCA is a class modeling method, class assignment is based on fit of the unknowns to the class models. This assignment allows the classification result that the unknown is none of the described classes, and has the advantage of providing the relative geometric portion of the newly classified object. This makes it possible to assess or quantitate the test sample in terms of external variables that are available for the training sets. 

Nevertheless, in most of the electronic tongue applications found in the literature, classification techniques like linear discriminant analysis (LDA) and partial least squares discriminant analysis (PLS-DA) have been used in place of more appropriate class-modeling methods. Moreover, in the few cases in which a class-modeling technique such as soft independent modeling of class analogy (SIMCA) is applied, attention is frequently focused only on its classification performance (e.g., correct classification rate). Use of such a restricted focus considerably underutilizes the significant characteristics of the class-modeling approach. 

Historically, SIMCA [43,44], proposed by Wold et al. in 1976, was the first class-modelling method introduced in the literature. Its key assumption is that the main systematic variability characterizing the samples from a category can be captured by a principal component model (see Chapter 4) of opportune dimensionality, built on training samples from that class. In detail, defining... 

The final class of methods that we shall consider for calculating the electrostatic compone of the solvation free energy are based upon the Poisson or the Poisson-Boltzmann equatior Ihese methods have been particularly useful for investigating the electrostatic properties biological macromolecules such as proteins and DNA. The solute is treated as a body of co stant low dielectric (usually between 2 and 4), and the solvent is modelled as a continuum high dielectric. The Poisson equation relates the variation in the potential (f> within a mediu of uniform dielectric constant e to the charge density p ... 

There are two main classes of loop modeling methods (1) the database search approaches, where a segment that fits on the anchor core regions is found in a database of all known protein structures [62,94], and (2) the conformational search approaches [95-97]. There are also methods that combine these two approaches [92,98,99]. 

Methods based on linear projection transform input data by projection on a linear hyperplane. Even though the projection is linear, these methods may result in either a linear or a nonlinear model depending on the nature of the basis functions. With reference to Eq. (6), the input-output model for this class of methods is represented as... 

Another class of methods of unidimensional minimization locates a point x near x, the value of the independent variable corresponding to the minimum of /(x), by extrapolation and interpolation using polynomial approximations as models of/(x). Both quadratic and cubic approximation have been proposed using function values only and using both function and derivative values. In functions where/ (x) is continuous, these methods are much more efficient than other methods and are now widely used to do line searches within multivariable optimizers. 

The various types of successful approaches can be classified into two groups empirical model calculations based on molecular force fields and quantum mechanical approximations. In the first class of methods experimental data are used to evaluate the parameters which appear in the model. The shape of the potential surfaces in turn is described by expressions which were found to be appropriate by semiclassicala> or quantum mechanical methods. Most calculations of this type are based upon the electrostatic model. Another more general approach, the "consistent force field method, was recently applied to the forces in hydrogen-bonded crystals 48> 49>. 

Whether the prediction scheme is a simple chart, a formula, or a complex numerical procedure, there are three basic elements that must be considered meteorology, source emissions, and atmospheric chemical interactions. Despite the diversity of methodologies available for relating emissions to ambient air quality, there are two basic types of models. Those based on a fundamental description of the physics and chemistry occurring in the atmosphere are classified as a priori approaches. Such methods normally incorporate a mathematical treatment of the meteorological and chemical processes and, in addition, utilize information about the distribution of source emissions. Another class of methods involves the use of a posteriori models in which empirical relationships are deduced from laboratory or atmospheric measurements. These models are usually quite simple and typically bear a close relationship to the actual data upon which they are based. The latter feature is a basic weakness. Because the models do not explicitly quantify the causal phenomena, they cannot be reliably extrapolated beyond the bounds of the data from which they were derived. As a result, a posteriori models are not ideally suited to the task of predicting the impacts of substantial changes in emissions. 

The similarity of samples can be evaluated by using geometrical constructs based on the standard deviation of the objects modeled by SIMCA. By enclosing classes in volume elements in descriptor space, the SIMCA method provides information about the existence of similarities among the members of the defined classes. Relations among samples, when visualized in this way, increase one s ability to formulate questions or hypotheses about the data being examined. The selection of variables on the basis of MPOW also provides clues as to how samples within a class are similar, and the derived class model describes how the objects are similar, with regard to the internal variation of these variables. 

Not every PRM is suitable for constructing spectral identification libraries. These are usually compiled by using supervised modeling methods, and unknown samples are identified with those classes they resemble most. 

Two class-modeling techniques have recently been introduced multivariate range modeling (MRM) and CAIMAN analogues modelling methods (CAMM). 

These efforts to improve the classification ability by correction of the original LDA model with the use of graphical means that search for a better discriminant line are the prelude to the use of classification methods with separate class models the bayesian analysis. 

The category correlations can be cancelled only when all the objects of the training set are in the same category, and the method is used as a class modelling technique. However, the bayesian analysis in ARTHUR-BACLASS has b n compared with the usual BA in classification problems about winra and olive oils and about the same classification and prediction abilities were observe for both methods. 

The SIMCA distances from two class models (or from the two models of the same category obtained by different methods) are reported in Coomans diagrams (Fig. 26) to show the results of modelling-classification analysis. 

The SIMCA method has been developed to overcome some of these limitations. The SIMCA model consists of a collection of PCA models with one for each class in the dataset. This is shown graphically in Figure 10. The four graphs show one model for each excipient. Note that these score plots have their origin at the center of the dataset, and the blue dashed line marks the 95% confidence limit calculated based upon the variability of the data. To use the SIMCA method, a PCA model is built for each class. These class models are built to optimize the description of a particular excipient. Thus, each model contains all the usual parts of a PCA model mean vector, scaling information, data preprocessing, etc., and they can have a different number of PCs, i.e., the number of PCs should be appropriate for the class dataset. In other words, each model is a fully independent PCA model. 

Chemical Class Model Compound Concentration (pg/L) Screening Concentration Testa Testa Method of Detectionb... 

---