Chiara Casolino

ITERATIVE PREDICTOR WEIGHTING (IPW) PLS: A TECHNIQUE FOR THE ELIMINATION OF USELESS PREDICTORS IN REGRESSION PROBLEMS

A new method for the elimination of useless predictors in multivariate regression problems is proposed. The method is based on the cyclic repetition of PLS regression. In each cycle the predictor importance (product of the absolute value of the regression coefficient and the standard deviation of the predictor) is computed, and in the next cycle the predictors are multiplied by their importance. The algorithm converges after 10–20 cycles. A reduced number of relevant predictors is retained in the final model, whose predictive ability is acceptable, frequently better than that of the model built with all the predictors. Results obtained on many real and simulated data are presented, and compared with those obtained from other techniques. Copyright

Characterisation of PDO olive oil Chianti Classico by non-selective (UV–visible, NIR and MIR spectroscopy) and selective (fatty acid composition) analytical techniques

An authentication study of the Italian PDO (protected designation of origin) extra virgin olive oil Chianti Classico was performed; UV-visible (UV-vis), Near-Infrared (NIR) and Mid-Infrared (MIR) spectroscopies were applied to a set of samples representative of the whole Chianti Classico production area. The non-selective signals (fingerprints) provided by the three spectroscopic techniques were utilised both individually and jointly, after fusion of the respective profile vectors, in order to build a model for the Chianti Classico PDO olive oil. Moreover, these results were compared with those obtained by the gas chromatographic determination of the fatty acids composition. In order to characterise the olive oils produced in the Chianti Classico PDO area, UNEQ (unequal class models) and SIMCA (soft independent modelling of class analogy) were employed both on the MIR, NIR and UV-vis spectra, individually and jointly, and on the fatty acid composition. Finally, PLS (partial least square) regression was applied on the UV-vis, NIR and MIR spectra, in order to predict the content of oleic and linoleic acids in the extra virgin olive oils. UNEQ, SIMCA and PLS were performed after selection of the relevant predictors, in order to increase the efficiency of both classification and regression models. The non-selective information obtained from UV-vis, NIR and MIR spectroscopy allowed to build reliable models for checking the authenticity of the Italian PDO extra virgin olive oil Chianti Classico.

Near infrared spectroscopy and class modelling techniques for the geographical authentication of Ligurian extra virgin olive oil

An authentic food is one which is what it purports to be. Food processors and consumers need to be assured that when they pay for a specific product, they are receiving exactly what they pay for. In this paper, a particular food authenticity study is considered: the classification of extra virgin olive oils from Liguria, a region in northern Italy, according to their geographical origin. One hundred and ninety five olive oil samples were analysed using a near infrared (NIR) instrument and the recorded spectra were used to build a class model for Ligurian olive oil. Different class modelling techniques were used, i.e. potential functions techniques (POTFUN), soft independent modelling of class analogy (SIMCA), unequal-quadratic discriminant analysis (UNEQ-QDA) and multivariate range modelling (MRM). In order to remove systematic variation in experimental data such as base-line and multiplicative scatter effects, an evaluation of different data pre-processing methods was performed. Ligurian olive oil was clearly differentiated from the other oils and the multivariate analysis allowed the construction of Liguria class models with good predictive ability, high sensitivity and sufficient specificity. The results obtained suggest that NIR and chemometrics are useful tools in the geographic traceability of olive oil.

Extraction of representative subsets by potential functions method and genetic algorithms

Abstract Two procedures are suggested to select a representative subset from a large data set. The first is based on the use of the estimate of the multivariate probability density distribution by means of the potential functions technique. The first object selected for the subset is that for which the probability density is larger. Then, the distribution is corrected, by subtraction of the contribution of the selected object multiplied by a selection factor. The second procedure uses genetic algorithms to individuate the subset that reproduces the variance–covariance matrix with the minimum error. Both methods meet the requirement to obtain a representative subset, but the results obtained with the method based on potential functions are generally more satisfactory in the case when the original set is not a random sample from an infinite population, but is the finite population itself. Several examples show how the extraction of a representative subset from a large data set can give some advantages in the use of representation techniques (i.e., eigenvector projection, non-linear maps, Kohonen maps) and in class modelling techniques.

The refinement of PLS models by iterative weighting of predictor variables and objects

The flexibility of PLS algorithm can be used to assign suitable weights to predictors or to objects or to both predictors and objects. Weights of predictors are obtained from the regression coefficients and the standard deviation. Weights of objects are obtained from the prediction residuals. By iterative weighting, the regression models are refined and a steady state is attained, where useless predictors and anomalous objects are cancelled, and a very economical model is obtained. The predictive ability and stability of this final model are better than those of the original model with all the available predictors and objects.