Development of long-short term memory neural network (LSTM) based crime prediction model

Abstract

Data mining is used to detect patterns in the dataset within the labelled data and also predicts future trends by unraveling the latten relationship existing within data that allow organizations to make informed decisions. Vanishing gradient problems within the data mining process serve a major constrain in crime prediction. This study adopts long short-term memory neural network (LSTM) that handles the vanishing gradient problem within the data mining process to predict future crime. The model was implemented as an intelligent system using Python. Also, the model was evaluated with 140,000 (one hundred and forty thousand) crime datasets from north-wale and north-Yorkshire police stations in the United Kingdom. The model was able to predict the crime rate for the supplied predicted year as against the dataset for the index year where it shows prediction accuracy of about 92.9%. The study recommends the full adoption of the system.