Table 1

Proportion of exposure predictions from CART and random forest models that agreed with the expert exposure estimate in the validation dataset (n=2248)

		Agreement (%)
Expert exposure metric	Number of jobs	CART, OH variables	CART, OH and module variables	Random Forest, OH variables	Random Forest, OH and module variables
Probability, binary
Negligible/low	1887	93	95	93	95
Medium/high	358	79	89	80	92
Overall	2245	92	94	92	94
Probability, ordinal
Negligible	1705	97	97	98	98
Low	182	32	43	23	32
Medium	73	7	21	8	14
High	285	68	85	72	85
Overall	2245	85	89	85	89
Intensity
None	1708*	98	97	96	97
Low	394	64	68	67	71
Medium	86	48	56	41	57
High	57	61	65	61	60
Overall	2245	87	89	88	90
Frequency
None	1757*	97	97	98	98
Low	209	32	52	26	34
Medium	141	12	39	13	35
High	141	57	63	59	65
Overall	2248†	83	87	84	86

*The number of unexposed jobs based on intensity and frequency exceeds the number of unexposed jobs based on the probability metric. This occurred because the estimated level of exposure intensity and frequency for some jobs did not exceed the minimum threshold (<0.25 µg m⁻³ REC and <0.25 h per week, respectively), even though a diesel exposure source was identified.
†There were three additional observations for the frequency metric; these observations were excluded from the probability and intensity analyses, because for those metrics an ‘unknown’ had been assigned.
CART, classification and regression tree; OH, occupational history; REC, respirable elemental carbon.