A Data-Driven Approach to Estimate Incident-Induced Delays Using Incomplete Probe Vehicle Data: Application to Safety Service Patrol Program Evaluation

Table 1

Comparison of performance metrics of different classification models.

Methods	Delay occurrence	Precision	Recall	F1 score	Accuracy (CR)	Delay occurrence classification (cases)		Model specification
						No	Yes
ANN	No delay	0.78	0.66	0.71	0.68	529^(1)	270^(2)	Hidden layers: 1,000, learning rate: 0.01, max iteration: 1,000
	Delay	0.57	0.71	0.63		152^(3)	365^(4)
SVC	No delay	0.68	0.92	0.78	0.69	735	64	Kernel: “Radial basis function (RBF)”
	Delay	0.72	0.32	0.45		349	168
NB	No delay	0.71	0.57	0.64	0.60	459	340	—
	Delay	0.49	0.64	0.56		186	331
KNN	No delay	0.70	0.88	0.78	0.70	707	92	Number of neighbors: 95
	Delay	0.70	0.41	0.52		303	214
RF	No delay	0.78	0.83	0.81	0.76	667	132	Number of estimators: 800
	Delay	0.71	0.64	0.68		186	331

⁽¹⁾True negative (TN), ⁽²⁾false positive (FP), ⁽³⁾false negative (FN), and ⁽⁴⁾true positive (TP).