GA-CatBoost-Weight Algorithm for Predicting Casualties in Terrorist Attacks: Addressing Data Imbalance and Enhancing Performance

Abstract

Terrorism poses a significant threat to international peace and stability. The ability to predict potential casualties resulting from terrorist attacks, based on specific attack characteristics, is vital for protecting the safety of innocent civilians. However, conventional data sampling methods struggle to effectively address the challenge of data imbalance in textual features. To tackle this issue, we introduce a novel algorithm, GA-CatBoost-Weight, designed for predicting whether terrorist attacks will lead to casualties among innocent civilians. Our approach begins with feature selection using the RF-RFE method, followed by leveraging the CatBoost algorithm to handle diverse modal features comprehensively and to mitigate data imbalance. Additionally, we employ Genetic Algorithm (GA) to finetune hyperparameters. Experimental validation has demonstrated the superior performance of our method, achieving a sensitivity of 92.68% and an F1 score of 90.99% with fewer iterations. To the best of our knowledge, our study is the pioneering research that applies CatBoost to address the prediction of terrorist attack outcomes.

Keywords

• terrorist attack prediction
• feature selection
• CatBoost
• sample imbalance