This paper describes an innovative way to optimize a multivariate classifier, in particular a Support Vector Machine algorithm, on a problem characterized by a biased training sample. This is possible thanks to the feedback of a signal-background template fit performed on a validation sample and included both in the optimization process and in the input variable selection. The procedure is applied to a real case of interest at hadron collider experiments: the reduction and the estimate of the multi-jet background in the \(W\to e \nu\) plus jets data sample collected by the CDF experiment. The training samples, partially derived from data and partially from simulation, are described in detail together with the input variables exploited for the classification. At present, the reached performance is superior to any other prescription applied to the same final state at hadron collider experiments.