Reti neurali e data mining per l’analisi della customer satisfaction: il caso della qualità della didattica nella Facoltà di Economia di Parma

The first aim of this paper is to present an original proposal on the measurement of customer satisfaction. The statistical method is based on two different types of neural networks: the self organizing maps and the radial basis function networks. The latter are implemented with orthogonal least squares selection of the basis functions, in order to avoid unstable results and computational problems in the subset selection algorithm due to multicollinearity in the input matrix. Orthogonal least squares are also much faster than forward selection in data mining, when the input matrix is large. Particularly attention is paid on the pre-processing and processing step (choice of the parameters and stability of the results) and on the definition of multivariate outliers. The case study presented throughout the paper is the measurement of the student satisfaction in the Faculty of Economics of the University of Parma and the quantification of the lag between perceived and expected quality.