Fig. 5

a Training process of EPA. The training set (“n” and “m” represent the number of data points and models respectively) was firstly augmented by m AAE models followed by training m DNN models. Then DNNs were used to re-predict the training set, generating n × m data points which were fed into RF model. Finally, the test set was predicted by the trained DNNs and RF model to evaluate the performance of EPA. b The performance of EPA compared with ENB, SmoteR and random under sampling (RUS). Statistical significance of the difference between the performance of EPA and ENB was determined by paired t-test. ns: p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001