Spatio-temporal data are data of events that occur in a certain place at a certain time, which have both temporal and spatial dimensions. “Crime” is an important example of spatio-temporal data, and crimes committed by people throughout history have been greatly influenced by temporal, spatial, relational, and environmental factors. Since it is an act that directly or indirectly leads to the deterioration of the social order, public authorities try to do their best to prevent crime throughout the human history.

 

Recent developments in artificial intelligence, machine learning and deep learning have led to emergence of many prediction systems in a wide range of domains. Revealing the temporal, spatial, relational, and environmental connections related to spatio-temporal data and using these connections in the development of forecasting systems make it possible to predict which type of an event will occur in which time & place. In this context, the development of crime prediction models is very important both in terms of being a hot topic and in terms of the potential big social impact if successful.

 

The concentration of certain crimes in certain locations (region, neighborhood, province, etc.) shows that spatial information has a dominant effect in crime prediction. Therefore, to model the crime events that take place in different places, there is a need for separate models that will use the features of each place effectively. On the other hand, it is possible that the crime events that take place in a certain region are related to the events that take place in different regions. For example, growing crime incidents in a certain region for economic reasons may be a sign that similar incidents will increase in the neighboring regions.

 

In several different customer cases with Predy, we provided significantly accurate crime prediction performance in real life operational conditions by modeling the temporal and spatial components of the related data as connected graphical networks, both within themselves and among each other.

 

The forecasting models developed with Predy has a structure in which forecast model of each location exchanges data with the models of the neighboring locations. Thus, neighboring prediction models can share prediction parameters or feature vectors with each other and use the data they receive from neighboring nodes to update their prediction models. Hence, Predy can produce high-accuracy crime predictions and the results it produces are used daily by the relevant security authorities operationally.