Geospatial Crime Hotspot Detection: A Robust Framework using Birch Clustering Optimal Parameter Tuning

Crime causes physical and mental damage. Several crime prevention measures have been developed by law enforcement officials since they realized how serious this problem is. These preventative measures are not strong enough to help lower crime rates because they are typically slow-paced and ineffectual. In this regard, machine learning community has started developing automated approaches for detecting crime hotspot, after performing a careful analysis of the crime trend incorporating geospatial, temporal, demographic, or other relevant information. In this research, we look at detecting crime hotspots using geospatial information of prior crime occurrences. We proposed BIRCH algorithm to detect high crime prone areas with four essential aspects: (1) PCA (Principle Component Analysis) has been used to minimize the dimensionality of crime data, (2) Silhouette score Elbow and Calinski Harabaz have been used to find the optimal number of cluster (3) utilized hyper-parameter tuning to choose the best hyperparameters for the BIRCH algorithm (4) applied BIRCH with the three aspects mentioned above. The results of the suggested framework were then contrasted with those of alternative clustering techniques, such as K-means, DBSCAN, and the agglomerative algorithm. We explored our approaches on the London Crime Dataset and found some fascinating results that can help reducing crime by helping people take the appropriate measures.