Variable Selection in Logistic Regression

In many fields, as for example in transportation system, in medicine, in chemiometry or in the evaluation of the Customer Satisfaction we deal with a dependent dichotomous character and with many explicative variables. In this situation the use of a linear regression model is not possible and the Logistic Regression Model is one of the possible solutions. Anyway, in many circumstances, one of the problems that can affect this model is the fact that there are few observations and many explicative variables. In this case the parameter estimation is computationally very expensive and the procedure can frequently lead to erroneous results. To solve this kind of problems, some possible solutions have been proposed in last years; in this paper we propose to use the Disco Coefficient to individuate the variables that can be significant for the Logistic Regression.