Riccardo Leardi

Genetic algorithms applied to feature selection in PLS regression: how and when to use them

Abstract Genetic algorithms (GA) are very useful in solving complex problems of optimization. The selection of the best subset of variables is surely one of them. In this paper, a new approach is proposed, and the positive and negative aspects of the application of GA in selecting variables for a partial least squares (PLS) model are taken into account. Finally, the analysis of the results obtained on several real data sets allows to find a rationale for a sensible application, showing that, if correctly applied, this technique almost always produces very good results.

Application of genetic algorithm–PLS for feature selection in spectral data sets

After suitable modifications, genetic algorithms can be a useful tool in the problem of wavelength selection in the case of a multivariate calibration performed by PLS. Unlike what happens with the majority of feature selection methods applied to spectral data, the variables selected by the algorithm often correspond to well‐defined and characteristic spectral regions instead of being single variables scattered throughout the spectrum. This leads to a model having a better predictive ability than the full‐spectrum model; furthermore, the analysis of the selected regions can be a valuable help in understanding which are the relevant parts of the spectra. After the presentation of the algorithm, several real cases are shown. Copyright

Variable selection for multivariate calibration using a genetic algorithm: prediction of additive concentrations in polymer films from Fourier transform-infrared spectral data

Abstract Variable selection using a genetic algorithm is combined with partial least squares (PLS) for the prediction of additive concentrations in polymer films using Fourier transform-infrared (FT-IR) spectral data. An approach using an iterative application of the genetic algorithm is proposed. This approach allows for all variables to be considered and at the same time minimizes the risk of overfitting. We demonstrate that the variables selected by the genetic algorithm are consistent with expert knowledge. This very exciting result is a convincing application that the algorithm can select correct variables in an automated fashion.

Genetic-algorithm-based wavelength selection in multicomponent spectrophotometric determination by PLS: application on copper and zinc mixture

Genetic algorithm (GA) is a suitable method for selecting wavelengths for PLS (partial least squares) calibration of mixtures with almost identical spectra without loss of prediction capacity using spectrophotometric method. The method is based on the development of the reaction between the analytes and Zincon at pH 9. A series of synthetic solution containing different concentrations of copper and zinc were used to check the prediction ability of the GA-PLS models. The RMSD for copper and zinc with GA and without GA were 0.0407 and 0.0865, 0.2147 and 0.3005, respectively. Calibration matrices were 0.05-1.8 and 0.05-1.5 mug ml(-1) for copper and zinc, respectively. This procedure allows the simultaneous determination of cited ions in natural, tap and waste waters good reliability of the determination was proved.

Chemometric analysis of Tuscan olive oils

Abstract Armanino, C., Leardi, R., Lanteri, S. and Modi, G., 1989. Chemometric analysis of Tuscan olive oils. Chemometrics and Intelligent Laboratory Systems, 5: 343–354. The chemical information (fatty acids, sterols, triterpenic alcohols) on 120 olive oil samples from Tuscany, Italy, collected in 88 different areas of production, was evaluated by display methods and cluster analysis. Inside this small region of varied orography, four groups of similar samples were detected and some relationships with the geographic profile were revealed by using a CAD package to produce effective geographical representations of variables, eigenvectors and clusters.

Genetic algorithms in chemometrics

This review covers the application of Genetic Algorithms (GAs) in Chemometrics. The first applications of GAs in chemistry date back to the 1970s, and in the last decades, they have been more and more frequently used to solve different kinds of problems, for example, when the objective functions do not possess properties such as continuity, differentiability, and so on. These algorithms maintain and manipulate a family, or population, of solutions and implement a “survival of the fittest” strategy in their search for better solutions. GAs are very useful in the optimization and variable selection in modeling and calibration because of the strong effect of the relationship between presence/absence of variables in a calibration model and the prediction ability of the model itself. This review is not a complete summary of the applications of GAs to chemometric problems; its goal is rather to show the researchers the main fields of application of GAs, together with providing a list of references on the subject. Copyright

Transfer of calibration function in near-infrared spectroscopy

Abstract A procedure for the transfer of the regression equation in near-infrared spectroscopy (NIRS), from a first instrument to a second instrument, is presented. The procedure uses partial least squares (PLS) regression twice: in the first step to compute the relationship between the spectra of transfer samples of the two instruments, and in the second step to compute the regression equation (relationship between chemical variables and spectral variables) of the first instrument. These two PLS steps are combined to predict the regression equation of the second instrument. Sometimes the PLS relationship between the two instruments is obtained from the principal components of the spectra of the two instruments. The procedure is applied to a set of 60 samples of soy flour, representative of the Italian soy production. 40 samples were used both as transfer samples and to compute the regression equation. 20 samples were used as evaluation set. Spectra were recorded with four different instruments, in four different laboratories. The result of the transfer procedure were evaluated by means of the standard error of prediction ( SEP ) with the predicted regression equation. Owing also to the great number of samples in the transfer set, and to the noise filtering effect of the twin PLS procedure, SEP with the predicted regression equation is not greater than that with the regression equation computed directly from the second instrument. The effect of some parameters, such as the number of PLS latent variables in the two steps, is also studied.

Three-mode principal component analysis of monitoring data from Venice lagoon

A data set obtained by 44 monthly determinations of 11 variables from 13 sampling sites in the Venice lagoon has been treated by three‐mode principal component analysis. The results show that the sampling sites are grouped according to their geographical location, following an inner–outer lagoon direction. In terms of sampling periods, a very strong seasonal effect has been detected, together with an almost linear decrease in nutrients (P and NO   3− ) and increase in eutrophication. Copyright

Detection of addition of barley to coffee using near infrared spectroscopy and chemometric techniques.

The current study presents an application of near infrared spectroscopy for identification and quantification of the fraudulent addition of barley in roasted and ground coffee samples. Nine different types of coffee including pure Arabica, Robusta and mixtures of them at different roasting degrees were blended with four types of barley. The blending degrees were between 2 and 20 wt% of barley. D-optimal design was applied to select 100 and 30 experiments to be used as calibration and test set, respectively. Partial least squares regression (PLS) was employed to build the models aimed at predicting the amounts of barley in coffee samples. In order to obtain simplified models, taking into account only informative regions of the spectral profiles, a genetic algorithm (GA) was applied. A completely independent external set was also used to test the model performances. The models showed excellent predictive ability with root mean square errors (RMSE) for the test and external set equal to 1.4% w/w and 0.8% w/w, respectively.

Genetic algorithm as a tool for selection of D-optimal design

Abstract Asymmetrical designs are now widely used in the industrial world. However, the traditional methods to construct them are very poor and difficult to handle. To construct experimental design in ‘non-standard’ occurrences, algorithmic methods, such as Fedorovs or Mitchells algorithms, are broadly used. But in case of high dimensionality, this class of algorithms cannot be used. In this paper, a new technique based on the genetic algorithm for constructing experimental designs is described.