Table 1.

Neural networks enable accurate performance prediction in flow-focusing droplet generation. The metrics are reported for a 20% test-set (training the models on 80% of the data and leaving 20% for the test-set).

Parameter	Regime	R2*	RMSE**	MAPE***	MAE****
Droplet diameter	Dripping	0.893  ±  0.029	13.1  ±  1.6 μm	11.2  ±  1.3%	9.9  ±  1.2 μm
Droplet diameter	Jetting	0.966  ±  0.010	8.2  ±  1.3 μm	4.8  ±  0.5%	5.9  ±  0.8 μm
Generation rate	Dripping	0.889  ±  0.026	31.7  ±  5.8 Hz	33.5  ±  4.2%	19.6  ±  2.7 Hz
Generation rate	Jetting	0.956  ±  0.009	21.9  ±  2.8 Hz	15.8  ±  2.9%	15.4  ±  2.1 Hz

*R² coefficient of determination, **Root mean square error, ***Mean absolute percentage error, ****Mean absolute error. The provided values are reported using the average plus-minus (±) the standard deviation for ten different training and testing sessions. For each session the test-set and train-set were randomly chosen.