Table 1.

Three representative deep multi-label classification models, TextCNN, TextRNN and TextRCNN for automatic ICD coding; all show signs of the overfitting issue. The algorithms obtained decent performance on the training data (top) but observed a large performance gap between training data and test data.

Models	TextCNN	TextRNN	TextRCNN
$\mathrm{F}{1}_{micro}$ on Training Data	0.695	0.789	0.624
$\mathrm{F}{1}_{micro}$ on Test Data	0.325	0.282	0.317