Non-linear methods

In recent years there has been much activity to devise methods for multivariate calibration that take non-linearities into account. Artificial neural networks (Chapter 44) are well suited for modelling non-linear behaviour and they have been applied with success in the field of multivariate calibration [47,48]. A drawback of neural net models is that interpretation and visualization of the model is difficult. Several non-linear variants of PCR and PLS regression have been proposed. Conceptually, the simplest approach towards introducing non-linearity in the regression model is to augment the set of predictor variables (jt, X2, ) with their respective squared terms (xf,. ..) and, optionally, their possible cross-product 

Hildrum, T. Isaksson, T. Naes and A. Tandberg, Near Infra-red Spectroscopy Bridging the Gap between Data Analysis and NIR Applications. Ellis Horwood, New York, 1992. 

van de Waterbeemd, ed., QSAR Chemometric Methods in Molecular Design. VCH, Weinheim, 1995. 

Naes and E. Risvik (Editors), Multivariate Analysis of Data in Sensory Science, Data Handling in Science and Technology Series. Elsevier, Amsterdam, 1996 P.K. Hopke and X.-H. Song, The chemical mass balance as a multivariate calibration problem. Chemom. Intell. Lab. Assist., 37 (1997) 5-14. 

Measurement, Calibration and Regression. Clarendon Press, Oxford, 1993. 

As mentioned at the beginning of this section, PCA lies at the heart of several analytical methods which will be discussed in later chapters. Some other features of PCs, such as their significance , are also discussed later in the book this section has been intended to illustrate the use of PCA as a linear dimension reduction method. 

For any given data set of points in N dimensions it is possible to calculate the distances between pairs of points by means of an equation such as that shown in eqn (4.2). 

This is the expression for the Euclidean distance where dy refers to the distance between points i and j in an A -dimensional space given by the summation of the differences of their coordinates in each dimension k= N). Different measures of distance may be used to characterize the 

Minimization of this error function results in a two-dimensional display of the data set in which the distances between points are such that they best represent the distances between points in A -space. The significance of the power term, p, will be discussed later in this section it serves to alter the emphasis on the relative importance of large versus small iV-space interpoint distances. 

A physical analogy of the process of NLM can be given by consideration of a three-dimensional object composed of a set of balls joined together by springs. If the object is pushed onto a flat surface and the tension in the springs allowed to equalize, the result is a two-dimensional representation of a three-dimensional object. The equalization of tension in the springs is equivalent to minimization of the error function in eqn 

---

Additionally, Breiman et al. [23] developed a methodology known as classification and regression trees (CART), in which the data set is split repeatedly and a binary tree is grown. The way the tree is built, leads to the selection of boundaries parallel to certain variable axes. With highly correlated data, this is not necessarily the best solution and non-linear methods or methods based on latent variables have been proposed to perform the splitting. A combination between PLS (as a feature reduction method — see Sections 33.2.8 and 33.3) and CART was described by... 

As an extension of perceptron-like networks MLF networks can be used for non-linear classification tasks. They can however also be used to model complex non-linear relationships between two related series of data, descriptor or independent variables (X matrix) and their associated predictor or dependent variables (Y matrix). Used as such they are an alternative for other numerical non-linear methods. Each row of the X-data table corresponds to an input or descriptor pattern. The corresponding row in the Y matrix is the associated desired output or solution pattern. A detailed description can be found in Refs. [9,10,12-18]. 

A final consideration about PCA is concerned with its use as a preprocessor of non-linear methods such as neural networks [22], The assumption of a normal distribution of the data requires all following analysis steps to adhere to this hypothesis. If positive results are sometimes achieved they have to be considered as serendipitous events. 

Other methods, which have not yet been used in chemical kinetics, include global or passive extrapolation (see Sect. 4.5.7), averaging methods, multistep, multiderivatives methods, exponential fitting and non-linear methods (see, for example, ref. 176 for references). 

Abstract Validation of analytical methods of well-characterised systems, such as are found in the pharmaceutical industry, is based on a series of experimental procedures to establish selectivity, sensitivity, repeatability, reproducibility, linearity of calibration, detection limit and limit of determination, and robustness. It is argued that these headings become more difficult to apply as the complexity of the analysis increases. Analysis of environmental samples is given as an example. Modern methods of analysis that use arrays of sensors challenge validation. The output may be a classification rather than a concentration of analyte, it may have been established by imprecise methods such as the responses of human taste panels, and the state space of possible responses is too large to cover in any experimental-design procedure. Moreover the process of data analysis may be done by non-linear methods such as neural networks. Validation of systems that rely on computer software is well established. 

Panaye A, Fan BT, Doucet JP, Yao XJ, Zhang RS, Liu MC, et al. Quantitative structure-toxicity relationships (QSTRs) A comparative study of various non linear methods. General regression neural network, radial basis function neural network and support vector machine in predicting toxicity of nitro- and cyano-aromatics to Tetrahymena pyriformis. SAR QSAR Environ Res 2006 17 75-91. 

Vyazovkin and Lesnikovich [42] have emphasized that the majority of NIK methods involve linearization of the appropriate rate equation, usually through a logarithmic transformation which distorts the Gaussian distribution of errors. Thus non-linear methods are preferable [89]. Militky and Sest [90] and Madarasz et al. [91] have outlined routine procedures for non-linear regression analysis of equation (5.5) above by transforming the relationship ... 

We analyze in this chapter recent developments in the area of ion-solid interactions, describing in particular the non-linear method to study the energy loss of ions in solids. We consider the non-perturbative scheme provided by quantum scattering theory, using self-consistent methods to... 

To test this property, we have condensed in Fig. 9 the results of numerous calculations using the present non-linear method [50], for ions with atomic numbers in the range 1 < < 92, and assuming always q = exit — sg-... 

The simplicity of chemical systems and the very complicated dynamics of chemical processes cause the mathematical models of chemical reactions to be an important area of applications of non-linear methods of mathematics, including catastrophe theory. 

More often the problem is non-linear and there is a choice of non-linear methods as illustrated in Figure 15. The most commonly used approach to solve a non-linear problem, such as the ExOR one, is an artificial neural network. 

Linearity of a method should be established or a series of standards selected for use with non-linear-method calibration. This can be checked by preparing and analyzing serial dilutions of aqueous reference standard solutions, quality control materials, enzyme solutions, or commercially available materials for demonstrating linearity (again, these are designed for use in human medicine) and comparing the determined values with the theoretical values calculated for the dilutions. The serial dilutions used for linearity checks can also help establish the analytical sensitivity when defined as the minimal detectable change from one concentration to another. 

Thus, the non-negativity of solution is not an established constraint in the theoretical foundation of linear methods. On the other hand, the empirically formulated non-linear methods [Eqs. (55-56)] effectively secure positive and stable solutions. Such a weakness of the rigorous linear methods indicates a possible inadequacy in criteria employed for formulating the optimum solutions. In Section 6 we discuss possible revisions in assumptions employed for accounting for random noise in inversions. For example, it will be shown that by using log-normal noise assumptions the non-negativity constraints can be imposed into inversion in a fashion consistent with the presented approach inasmuch as one considers the solution as a noise optimization procedure. 

Ho YS. Second-order kinetic model for the sorption of cadmium onto tree fern a comparison of linear and non-linear methods. Water Res 2006 40 119-25. 

In compositional data, as pointed out Berrueta et al. (2007), the main problem is class overlap, but with a suitable feature selection and adequate sample size, good classification performances can be achieved. In general, non-linear methods such as ANN or SVM are rarely needed and most classification problems can be solved using linear techniques (LDA, CVA, PLS DA). 

However, the last traces of doubt have to be eliminated by measurements made at the bridgesite in this way we also hope to verify the non-linear methods of calculating the longitudinal forces and their superposition. 

Non-linear method Rate equations (2) and (3) may be solved numerically using a computer program. This ap>proach has been used for several epoxy/aUphatic diamine and good fits of epoxy and primary amine conversions were obtained (Figure 11) (M. Gonzalez et al., 2003). 

Hypertext refers to a non-sequential, non-linear method for organizing and displaying text It was designed to enable the reader to access information from a text in ways that are most meaningful (Nelson, 1981), based upon the assumption that the organization that the reader imposes on a text is more meaningful than that preferred by the author. 

To build a system capable of predicting characteristics of interest from measured values, we need to form a model which relates the measurements to the physical effect of interest. This model can then be applied to future measurements to predict the effect. In mathematical terms, the objective is to find the multidimensional function,/, such that y =f x), where jc is a new multivariate reading and y is the effect to be predicted. For the present, we shall assume that / is a linear function, for simplicity. Artificial neural networks (ANNs), a non-linear method, will be discussed later. 

There are several cases where NMR spectroscopy has been used to investigate copolymers which deviate from the terminal model for copolymerisation (see also chapter 3). For example, Hill and co-workers [23, 24] have examined sequence distributions in a number of low conversion styrene/acrylonitrile (S/A) copolymers using carbon-13 NMR spectroscopy. Previous studies on this copolymer system, based on examination of the variation of copolymer composition with monomer feed ratio, indicated significant deviation from the terminal model. In order to explain this deviation, propagation conforming to the penultimate (second-order Markov) and antepenultimate (third-order Markov) models had been proposed [25-27]. Others had invoked the complex participation model as the cause of deviation [28]. From their own copolymer/comonomer composition data. Hill et al [23] obtained best-fit reactivity ratios for the terminal, penultimate, and the complex participation models using non-linear methods. After application of the statistical F-test, they rejected the terminal model as an inadequate description of the data in comparison to the other two models. However, they were unable to discriminate between the penultimate and complex participation models. Attention was therefore turned to the sequence distribution of the polymer. 

Optimization of performance characteristics linear and non-linear methods... 

The non-linear methods have been used in combination with the methods of response surface design which generated the objective function based on experimental data. 

Non-linear methods are used not only for the optimization as illustrated above but also in regression analysis when fitting functions which are nonlinear with respect to their coefficients. For example, an application of the least squares method for the estimation of coefficients a and b of the function y = a(l-exp(-bx)) leads to NLP. In Section 1.4.3 NLP has been used for the estimation of unknown parameters of a fibre migration model. 

---