Quantitative Analysis and Prediction of Global Terrorist Attacks Based on Machine Learning

Table 5

Comparison of 10-fold cross-validation and hold-out methods (terror attack frequency ≥ 500).

Metrics		Data split verification method
		Hold-out method	10-fold cross-validation method
			Mean	Max	Min
Accuracy	Decision trees	0.958647	0.956026	0.964387	0.949943
	Bagging	0.931989	0.933528	0.936764	0.927569
	Random forests	0.968216	0.965974	0.968744	0.962647
	ExtraTrees	0.959501	0.959406	0.962400	0.956011
	XGBoost	0.971634	0.967778	0.970828	0.963216
Precision	Decision trees	0.928242	0.929096	0.943624	0.918073
	Bagging	0.932152	0.927242	0.938100	0.911614
	Random forests	0.957727	0.955594	0.963753	0.947774
	ExtraTrees	0.942159	0.941508	0.945737	0.935197
	XGBoost	0.957246	0.952817	0.956800	0.943972
Recall	Decision trees	0.929554	0.931822	0.944081	0.923149
	Bagging	0.858106	0.862504	0.871233	0.854989
	Random forests	0.934140	0.934952	0.941422	0.929029
	ExtraTrees	0.926234	0.928944	0.934792	0.923533
	XGBoost	0.944904	0.942287	0.950476	0.933696
F1 score	Decision trees	0.928603	0.930063	0.943702	0.920590
	Bagging	0.875151	0.875903	0.886239	0.866793
	Random forests	0.942883	0.942658	0.949366	0.935658
	ExtraTrees	0.932798	0.933608	0.937390	0.927087
	XGBoost	0.950011	0.946542	0.953123	0.937474