Spatially correlated device activation is a typical feature of the Internet of Things (IoT). This motivates the development of channel scheduling (CS) methods that mitigate device collisions efficiently in such scenarios, which constitutes the scope of this work. Specifically, we present a quadratic program (QP) formulation for the CS problem considering the joint activation probabilities among devices. This formulation allows the devices to stochastically select the transmit channels, thus, leading to a soft-clustering approach. We prove that the optimal QP solution can only be attained when it is transformed into a hard-clustering problem, leading to a pure integer QP, which we transform into a pure integer linear program (PILP). We leverage the branch-and-cut (B&C) algorithm to solve PILP optimally. Due to the high computational cost of B&C, we resort to some sub-optimal clustering methods with low computational costs to tackle the clustering problem in CS. Our findings demonstrate that the CS strategy, sourced from B&C, significantly outperforms those derived from sub-optimal clustering methods, even amidst increased device correlation.