We present a novel deterministic model that is capable of predicting particle-to-particle force and torque fluctuations in a fixed bed of randomly distributed monodisperse spheres. First, we generate our dataset by performing particle-resolved direct numerical simulations (PR-DNS) of arrays of stationary spheres in moderately inertial regimes with a Reynolds number range of $2 \leq \textit {Re} \leq 150$ and a solid volume fraction range of $0.1 \leq \phi \leq 0.4$. The key idea exploited by our model is that, while the arrangement of neighbours around each particle is uniform and random, conditioning forces or torques exerted on a reference sphere to specific ranges of values results in the emergence of significantly non-uniform distributions of neighbouring particles. Based on probabilistic arguments, we take advantage of the statistical information extracted from PR-DNS to construct force/torque-conditioned probability distribution maps, which are ultimately used as basis functions for regression. Given the locations of surrounding particles as input to the model, our results demonstrate that the present probability-driven framework is capable of predicting up to 85 % of the actual observed force and torque variation in the best cases. Since the precise location of each particle is known in an Eulerian–Lagrangian (EL) simulation, our model would be able to estimate the unresolved subgrid force and torque fluctuations reasonably well, and thereby considerably enhance the fidelity of EL simulations via improved interphase coupling.