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

Table 12

A performance comparison of our method with previous research, regarding single-label and multilabel identification.

Studies	Algorithm	HF accuracy	AC accuracy	Remark
Shi et al. [4]	Naive Bayes	73.2%	81.1%	This study targets HF and AC only
	Hoeffding tree	74.9%	87.0%
	OzaBagADWIN	76.5%	88.3%
Our study	LSTM without fine-tuned language model	84.8%	85.1%	Our study achieves a better result regarding HF and can identify four more factors
	LSTM with fine-tuned language model	88.1%	87.0%
Studies	Algorithm	Hamming loss	score	Remark
Robinson [13]	Latent semantic analysis	0.269	0.409	Impractically targeting 16 factors
Our study	LSTM without fine-tuned language model	0.135	0.628	Our study feasibly targets the six most frequent factors with promising results achieved
	LSTM with the fine-tuned language model	0.091	0.763

The advantage of the deep learning methods over traditional machine learning methods is clearly shown.