An Alternative Method for Traffic Accident Severity Prediction: Using Deep Forests Algorithm

Table 4

The predictive performance of different accident categories.

	Recall			False alarm rates			F1 score			ROC
	1	2	3	1	2	3	1	2	3	1	2	3
Deep Forests	0.93	0.82	1.00	0.17	0.09	0.01	0.88	0.86	1.00	0.91	0.88	1.00
RFs	0.91	0.77	1.00	0.19	0.10	0.01	0.86	0.83	1.00	0.90	0.86	1.00
XGboost	0.83	0.66	1.00	0.27	0.20	0.02	0.78	0.72	0.99	0.84	0.79	0.99
LightGBM	0.84	0.63	1.00	0.29	0.20	0.02	0.77	0.71	0.99	0.84	0.78	1.00
Decision tree	0.68	0.76	1.00	0.23	0.28	0.05	0.77	0.72	0.95	0.78	0.80	0.98
KNN	0.66	0.64	1.00	0.32	0.33	0.07	0.67	0.66	0.97	0.75	0.75	0.98
DNN	0.80	0.07	0.70	0.52	0.51	0.38	0.60	0.13	0.66	0.56	0.43	0.56