Comparaison de la prédictivité d'un réseau de neurones à rétropropagation avec celles des méthodes de régression linéaire, logistique et AID pour le calcul des scores en marketing direct

In comparison with other statistical methods (MCO, logistic regression, disciminant analysis, AID), the advantages of a neural network with retropropagation are numerous and well known (non linear effects, distribution free variables, low sensibility to outlyers or missing variables). However, implementation and efficiency have not yet received a strong interest. The paper reviews comparative analyses and presents the results obtained in predicting a behavior in Fund raising.