Spatio-temporal prediction of terrorist attacks based on GCN-LSTM

Abstract

Terrorist attacks represent a significant threat to national order, social stability, and economic security. Accurate prediction of such attacks is a critical task for casualty reduction, enhanced decision-making, and optimal resource distribution in counter-terrorism efforts. This paper introduces an innovative spatio-temporal fusion framework that combines graph convolutional network (GCN) with long short-term memory (LSTM) models. By capturing and merging spatio-temporal features from relevant events, the proposed GCN-LSTM model achieves remarkable accuracy in predicting terrorist attacks. The experimental results demonstrate outstanding performance, with the model attaining minimal RMSE and MAE values of 0.037 and 0.031, respectively, surpassing all baseline models (LSTM, GCN, and CNN-LSTM-Transformer). Through its effective interpretation of complex spatio-temporal patterns underlying terrorist attacks, our model substantially enhances the predictive accuracy across diverse time horizons. These findings carry crucial implications for enhancing counter-terrorism strategies.