An architecture for emergency event prediction using LSTM recurrent neural networks

Emergency event prediction is a crucial topic since the events could involve human injuries or even deaths. Many countries record a considerable number of emergency events (EVs) that are caused by a variety of incidents such as murder and robbery. Emergency response systems based on more accurate EV prediction can help to allocate the required resources and resolve the emergencies through more rapid and effective risk management. Most real-time EV prediction systems are based on traditional time series analysis techniques such as moving average or autoregressive integrated moving average (ARIMA) models. To improve the accuracy of EV prediction, we propose a new architecture for EV prediction based on recurrent neural networks (RNN), specifically a long short-term memory (LSTM) architecture. A comparative analysis is presented to show the effectiveness of the proposed architecture compared to traditional time series analysis and machine learning methods through the evaluation of historical EV data provided by the national police of Guatemala.