In this research, we investigated the influence of a Fully Connected Deep Neural Network (FCN) for signal-to-background classification on a sample of Vector Boson Fusion (VBF) Higgs bosons decaying into b-quark pairs. The FCN improves the identification of the signal events overwhelmed by the QCD background. However, the selection of the signal efficiency Working Point has a sculpting effect on the background distribution of the invariant mass of the tagged jets. This condition is generated by the algorithm correlation with the Higgs boson mass. In fact, due to the input features non-linear dependence on the tagged jets' invariant mass, the algorithm learns that the signal event mass of the b-jet pair is close to the Higgs boson mass. Therefore the background events that have similar b-jets mass are misidentified as signal. In this paper, the correlation impact has been studied. Moreover, two different decorrelation approaches have been tested on Monte Carlo datasets self-produced.