In ecology, occupancy data are a contaminated binary response that is used to map the presence or absence of a species. Models for occupancy data are used to estimate the occurrence of a species, where the true presence of a species is a function of a spatially varying process. In the standard spatial occupancy model, most researchers assume that the spatial component is a Gaussian process. This assumption leads to an inability to identify non-traditional spatial dependence such as discontinuities and abrupt transitions which are common in ecological data. Bayesian machine learning techniques have the potential to identify non-traditional spatial structure, but these technologies do not account for contamination in the binary response. We embed Bayesian machine learning methods into the hierarchical occupancy model to account for non-traditional spatial dependence and contamination in the binary response. We conduct a simulation experiment by selecting a few commonly encountered cases of traditional and nontraditional spatial dependencies in ecology and include an application of our method using data on Thomson's gazelle in Tanzania.