Citation

Luo L, Qi C. Safety Sci. 2021; 144: e105442.

Copyright

(Copyright © 2021, Elsevier Publishing)

DOI

10.1016/j.ssci.2021.105442

PMID

unavailable

Abstract

The factors that may impact the risk of terrorist attacks are numerous and interrelated in a complex manner. This complexity makes the prediction of terrorist attacks challenging and leads to information redundancy and the obscuring of critical points. This paper aims at identifying crucial indicators from the perspective of predicting the risk of terrorist attacks. Both root cause and incident level factors are taken into account, which are qualified using 28 indicators. A random forest (RF) model is established to predict terrorist attack risk, and the prediction performance is recorded as the baseline result in terms of MAE,MSE, and R2. A recursive feature elimination method utilizing random forest kernels (RF-RFE) is proposed to identify crucial ones from the 28 initial indicators. The RF-RFE process gradually eliminates the least important indicators and compares the corresponding prediction performance with the baseline result. The prediction performance is relatively stable until the number of input indicators is reduced from 28 to less than 8. The indicators that make up the input set at the hedging point with eight indicators are considered as the most important ones, including Human_loss, GDPGrowth, MilitaryExpenditure, PopulationGrowth, Population(lg.), Unemployment, UrbanPopulationGrowth, InternalConflict, etc. The identified crucial indicators indicate that foresight and preemptive measures should be taken not only for specific intelligence and response operations, but also to improve the underlying government stability, economic quality, and other basic elements of citizens' lives.

Language: en

Keywords

Crucial indicator identification; Random forest; Recursive feature elimination; Terrorist attack prediction