The Standard Model of particle physics (SM) describes all known matter and its interactions - with the exception of gravity - to very high precision. The picture is not complete though, driving searches for physics Beyond the Standard Model (BSM). To find hints for BSM physics high precision testing of the predictions provided by the SM are required. One of the areas of interest in this regard is the study of rare decays, denoting decay processes with very low branching ratio predictions in the SM, but increased values in BSM theories. Rare decays accessible at the Belle experiment, which is located at the asymmetric energy electron-positron collider KEKB in Tsukuba, Japan, are leptonic decays of B_s^0 mesons, e.g. B_s^0 \to \tau \tau or B_s^0 \to \ell \tau. To circumvent difficulties in the reconstruction of these decay modes arising from not detected neutrinos in the final state, a method called hadronic tagging is used. With this technique one of the two B_s^0 mesons created at an electron-positron collision event is reconstructed in a full hadronic final state. To increase the reconstruction efficiencies of these hadronic tags compared to plain cut based selection, the inclusion of artificial neural networks into the reconstruction algorithms has proven effective in prior projects. This is due to the improved signal and background identification provided by neural networks, which are multivariate analysis tools mainly developed for machine learning purposes. Within this thesis the incorporation of 104 optimized neural networks into the hadronic tag, which was developed by the Vienna Belle Group, has been realised. In its finalized version the tag reaches a reconstruction efficiency of epsilon = 0.042% including 5416 full hadronic decay chains.