Small Area Crime Prediction using Deep Neural Net

Abstract

In this work, we have leveraged the power of Deep Neural Networks to make day to day crime count prediction in a small area of the city boundaries. We use Chicago city data to make the predictions. The crime data is augmented with weather, transportation and census data. We also study the effect of these augmented data on the accuracy of the model. We split the crime counts into 5 bins based on their counts. The model then predicts the most probable crime bin for each of the small areas on a day to day basis. The model is also trained on the variations in temporal and spatial characteristics of the crime prediction. The outcomes of the experiments demonstrate that the Deep Neural Networks are very effective in predicting crime count of a small geographical area.