Identifying Incident Causal Factors to Improve Aviation Transportation Safety: Proposing a Deep Learning Approach

Table 9

After applying the threshold, the model’s overall performance in terms of micro- score is improved, especially for rarer factors, as precision and recall become more balanced.

	Probability threshold = 0			Probability threshold = 0.55			score improvement
	Precision	Recall	Micro-	Precision	Recall	Micro-	Percentage
HF	0.306	0.92	0.502	0.373	0.84	0.516	+2.7%
AC	0.689	0.82	0.748	0.664	0.89	0.761	+1.7%
CP	0.756	0.31	0.440	0.778	0.42	0.545	+23.9%
PR	0.868	0.33	0.478	0.783	0.47	0.588	+23.0%
WE	0.882	0.59	0.706	0.855	0.59	0.702	− 0.5%
AP	0.971	0.35	0.514	0.83	0.47	0.596	+16.0%