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

Table 6

The summary of the incident reports and their label distribution in the training set before and after data oversampling, as well as validation and test sets.

	Original	Train (oversampled)	Validation	Test
Human factors (HF)	87356 (62.8%)	87356 (25.4%)	10941 (64.0%)	16145 (63.4%)
Aircraft (AC)	32690 (23.5%)	65380 (19.0%)	3823 (22.4%)	6620 (26.0%)
Company policy (CP)	5335 (3.8%)	53350 (15.5%)	635 (3.7%)	1047 (4.1%)
Procedure (PR)	5321 (3.8%)	53210 (15.4%)	645 (3.7%)	1004 (4.0%)
Weather (WE)	4979 (3.6%)	49790 (14.5%)	623 (3.7%)	952 (3.7%)
Airport (AP)	3424 (2.5%)	34240 (10.0%)	428 (2.4%)	643 (2.5%)
Total	139105 (100%)	343326 (100%)	17095 (100%)	25451 (100%)

Validation and test data are maintained as imbalanced as the original training set to truly represent the data sample distribution.