• artificial intelligence
• classification
• crime prediction
• real world datasets
• regression

Crime events in cities around the world have been proven to be unevenly distributed in space and time.
The first crime dataset that was analysed in detail contained over half a million crime records that were registered in the city of Boston, Massachusett. This dataset was explicitly used to understand the possible directions that could be followed to obtain interesting classification results. Next, a Swiss crime dataset was analysed. This dataset was much smaller compared to the previous one. After the preprocessing, many different types of analyses were made: the first was a regression analysis, which proved to obtain results that were of not very high quality. After this, various classification tasks (i.e. tasks with 2/3 classes, tasks that also considered exogenous data or only crimes that were committed during a specific part of the year) were described. The models that were used in these cases were to obtain very interesting results under specific circumstances. Finally, the classification models were compared to a state-of-the-art classification model, the Near Repeat prediction system, which was implemented in a way that could easily adapt to the structure of the given datasets.
The interesting results that were obtained in this thesis could certainly be used as a solid starting point for many other future analyses regarding the fascinating world of crime prediction.