We have tested brittleness prediction by integrating well and 3D seismic data using machine learning (proximal support vector machine algorithm) for Lower Paleozoic shales from the Baltic Basin in northern Poland. The workflow allowed for differentiation of the brittle and ductile zones of the thin shale layers, as well as mapping of the marly formation (fracture barrier) with superior resolution as compared with the resolution of the original input seismic data. The important part of the success was the appropriate definition of the mineralogical brittleness index (BI) tailored to the local geologic conditions. The obtained BI volume outlines the more and less brittle zones in the lower shale unit, i.e., the Sasino Formation, as well as the overlaying unit with high values of BI (Prabuty marls). The mechanical BI based on the Young’s modulus and Poisson’s ratio did not deliver the desired brittleness characterization of the formations of interest, which confirms the weakness of estimating BI using the above geomechanical measurements alone. The weak point of the reported analysis is the small number of available wells, which makes the prediction’s statistics unsatisfactory.