Unsupervised Domain Adaptation for Crime Risk Prediction Across Cities

Crime risk prediction is crucial for city safety and residents’ life quality. However, without labeled data, it challenging to predict crime in cities. Due municipal regulations maintenance costs, not trivial many cities collect high-quality data. In particular, some have lots of data while others may few. It has been possible develop a model by learning knowledge from with abundant Nevertheless, the inconsistency relevant context between exacerbates difficulty this task. To end, article proposes an effective unsupervised domain adaptation (UDAC) across addressing contexts’ issue. More specifically, we first identify several similar source grids each target grid. Based on these grids, then construct auxiliary contexts city, make consistent two A dense convolutional network designed learn high-level representations accurate simultaneously domain-invariant features adaptation. The effectiveness our verified through extensive experiments using three real-world datasets.