Fig. 4.

Reliability of model prediction. Validation performances, measured using F1 scores (a) and MCC (b), in predicting sex and age under balanced and confounded scenarios. According to data generation date, 27 libraries from the former nine batches were used as training set, while the remaining 18 libraries from the latter six batches were used as validation set. Prediction models were constructed based on training set using five machine-learning algorithms, including model averaged neural network (avNNet), support vector machine (SVM), random forest (RF), generalized partial least squares (GPLS), and linear algorithm BstLm. The models were then validated using the validation set and calculated the performances