We apply binary logistic regression to correlate fracture shear-slip criticality to hydraulic conductivity using data from four deep scientific boreholes in fractured crystalline rocks. In each borehole, an optimized decision boundary is obtained by maximizing the joint probability of classifying all fractures in consideration as critical or not. All four cases feature an optimized decision boundary close to the empirical rock friction (μ = 0.6), corroborating the applicability of laboratory-derived friction coefficients to faults in situ. Utilizing this statistical technique, we demonstrate that one can determine the in situ stress orientation and relative magnitude based only on whether fractures of varied orientations are hydraulically conductive, or not. The stress inversion results are consistent with independent stress measurements in each of the four case studies.