Geospatial and Temporal Modeling of Crime Rates Using Neural Networks

Crime forecasting has become an interesting concept of data science techniques like machine learning. In this research paper, we explore various architectures and algorithms of Neural Network (NN) models for predicting crime rates based on spatial and temporal crime data. The models are trained and tested on the dataset of crimes committed on a certain day of week, with geographic coordinates and generate a prediction or forecast of crimes to be committed across the region. The Long Short-Term Memory (LSTM) model of each of the algorithms to be explored achieves the highest accuracy. This research attempts to demonstrate the capability and reliability of NNs for crime forecasting based on the features already stated. This model could be of great value to law enforcement with resource allocation and strategic planning.