The current study examines whether the predictive modelling of crime can be applied consistently across different urban settings. An ensemble network was applied to generate crime predictions regarding three specific urban settings, for which both crime and supporting data were employed spanning the period from 2012-2016. For each setting, prediction performance measures were calculated and compared per crime type. The results indicate that relatively better and consistent performance measures were achieved for a larger and denser urban setting (setting C), while for a smaller urban setting (setting B), performance measures suggested that the model was overpredicting. For the urban setting with a more intermediate size (setting A), prediction performance was average compared to the other settings, yet for aggressive theft, relatively poor performance measures were achieved. Future research should consider to predict crime across urban and rural settings. Limitations of this study are furthermore discussed.