Time Series Crime Prediction Using a Federated Machine Learning Model

Crimes are a common social problem affecting quality of life. With an increase in the number of crimes, it is necessary to build a model to predict the number of crimes that might occur in a certain period, determine the characteristicsof a person who might commit a certain crime, and identify places where a certain crime might occur. Data privacy is the main challenge that organizations face when building this typeof predictive model. Federated learning (FL) is a promising approach that overcomes data security and privacy challenges, asit enables organizations to build a machine learning model based on distributed datasets without sharing raw data or violatingdata privacy. In this paper, we proposed a federated long short- term memory (LSTM) machine learning model and a traditional LSTM machine learning model by using TensorFlow Federated (TFF) and the Keras API to predict the number of crimes. During our experiment, we applied the proposed models on the Boston crime dataset. We attempted to change the proposed model's parameters to obtain minimum loss and maximum accuracy. Finally, we compared the federated LSTM model with the traditional LSTM model and found that the federated LSTMmodel resulted in lower loss, better accuracy, and higher trainingtime than the traditional LSTM model.