LSTM-Based Predictive Modeling for Crime Rate Forecasting Across Regions

Abstract— Urban and regional safety always matters and attracts so much interest up on crime and safety level predictions. Accuracy and presicion in the crime rate and safety level scores relate to data collected from previous occasions which remains very sensitive to time. The data dependency on time requires spacial care in handling and processing. Recurrent neural networks (RNN) have been used as alternative machine learning approach to statistical models in handling data with time sensitivity properties. Recently, long short term memory (LSTM) model has been used successfully in prediction, especially with timesensitive data. In this study, crime rate prediction for city of Bristol is investigated and analysed using LSTM models in comparison to well-know machine learning and statistical approaches, where LSTM model is found significantly outperforming. Keywords: Time Series Data, Long Short Term Memory, Crime Rate Prediction.