Medical Text Classification Using Hybrid Deep Learning Models with Multihead Attention

Table 8

Analysis with different convolution kernel conditions for the proposed hybrid BiGRU model.

Kernel filter sizes	Dataset	Accuracy (%)	Precision (%)	Recall (%)	F1-score (%)
[1, 3]	Hallmarks	71.85	61.79	60.70	61.24
	AIM	93.27	89.36	85.29	87.27
[1, 3, 5]	Hallmarks	73.88	69.54	68.67	69.10
	AIM	95.29	89.88	84.13	86.90
[1, 3, 5, 7]	Hallmarks	72.52	69.72	65.99	67.80
	AIM	82.79	76.91	73.78	75.31
[1, 3, 5, 7, 9]	Hallmarks	70.48	67.72	63.85	65.72
	AIM	58.39	51.84	57.18	54.37