Identifying and Labeling Potentially Risky Driving: A Multistage Process Using Real-World Driving Data

Table 6

Cross-classifying potentially risky driving behaviors.

		Dataset labeled
		April 1 (%)	April 2 (%)	April 4 (%)	April 5 (%)	April 6 (%)	April 7 (%)
Random forest models	April 1, 2013		49.1	65.7	47.1	46.6	52.8
	April 2, 2013	52.0		51.8	69.2	67.6	72.9
	April 4, 2013	59.3	49.2		47.7	50.0	57.0
	April 5, 2013	56.3	73.6	56.4		72.6	80.2
	April 6, 2013	50.6	69.0	54.4	69.4		73.4
	April 7, 2013	50.4	65.7	53.7	68.8	66.8

Percentages represent the proportion of the originally labeled observations (by the same-day model) that the cross-day model also identified. We note that all cross-classifications labeled a similar proportion of each dataset as potentially risky (∼5–10%).