| dc.description.abstract | As technology advances at an unprecedented pace, law enforcement faces the challenges of keeping up to date with cutting-edge methods to develop effective crime prevention practices. Using crime mapping and algorithms such as random forest to inform the development of crime prevention policies has the potential to reduce instances of violent crimes in Oslo significantly. The interest in predictive policing is increasing, but when it comes to new technologies that can help explore crime patterns, supervised machine learning models are empirically tested in Norway to a limited extent. Since the risk of biased assessments based on crime predictions can increase when technologies are not adequately understood and the applied input data quality could be better, expanding our knowledge in this field is crucial. This study evaluates the predictive performance of random forest models forecasting violent crimes at three different spatial levels in Oslo, Norway. Data from the Norwegian police crime registry and environmental features, including urban data and weather data, are used to enhance the prediction performance of the algorithm. Findings showed that random forest could predict violent crimes with up to 80 per cent accuracy. Here, the location and spatial time lags of violent crimes in Oslo were significant predictors of future crimes, as were environmental features such as minor roads, residential areas, and forests. These results suggest that violent crimes in Oslo exhibit spatiotemporal dependencies, which can increase the risk of near-repeat offending and contribute to further occurrences of violent crimes. The study concludes that using random forest algorithms in crime mapping is a highly accurate predictive model for law enforcement in Oslo. Still, there are some critical challenges that technological advancements may present for the implementation of new policies. Based on the empirical findings and a more comprehensive discussion of the effectiveness, limitations, and ethical implications of the approach, this study hopes to contribute to the current discourse on the responsible and effective adoption of data-driven strategies in crime prevention, acknowledging the need for adaptability and continuous learning in the face of ever-evolving technology. | eng |