AIST: An Interpretable Attention-based Deep Learning Model for Crime Prediction

Accuracy and interpretability are two essential properties for a crime prediction model. Accurate prediction of future crime occurrences along with the reason behind a prediction would allow us to plan the crime prevention steps accordingly. The key challenge in developing the model is to capture the non-linear and dynamic spatial dependency and temporal patterns of a specific crime category, while keeping the underlying structure of the model interpretable. In this article, we develop AIST, an Attention-based Interpretable Spatio Temporal Network for crime prediction. AIST models the dynamic spatio-temporal correlations for a crime category based on past crime occurrences, external features (e.g., traffic flow and point of interest information) and recurring trends of crime. Extensive experiments show that AIST outperforms the state-of-the-art techniques in terms of accuracy (e.g., AIST shows a decrease of 4.1% on mean average error and 7.45% on mean square error for the Chicago 2019 crime dataset) and interpretability.1