In the current security system, police always try to find the criminals after the crime has happened. For the elimination of these crimes, the government is always looking to adopt some decent and reliable strategy and a sustainable e-governance Information System. The prediction of crime would definitely be helpful in this case, especially with the details of time and space. For this purpose, research is done based on the spatial and temporal data for urban crime. The reason and requirement for choosing these spatio-temporal data is that crime mostly happens in a pattern, just like there are more theft cases at night rather than during the day. Similarly, residential areas on the outer part of the city are more likely to report theft cases.

So, the research idea is to observe the spatial and temporal data and come up with the hotspot areas and also to train ML models on these data to get the prediction of crime type with respect to location and time. As part of the experiments, I have followed the basic approach of training the weak learners like Random Forest, k-Nearest Neighbour, and MultiLayer Perceptron and refined them with Adaboost and GradientBoost boosting algorithms. Also, as part of a different approach, I have implemented a 3D CNN model on different crime datasets. The outcome of these experiments can be seen in this report.