Fig. 4

Feature selection by using \(\mathcal {L}1\)-logistic regression in human. Total data was randomly separated into 80% for training the model and 20% for the calculation of accuracy (blue dashed line, left y-axis). On the x-axis, C indicates the inverse of regularization strength. As C is increased, the number of features with non-zero coefficients (right y-axis) is increased and the model becomes more complicated. The black dashed line shows the final model chosen in this study, and outputs 15 features with non-zero coefficients. These features were ranked by the absolute value of coefficient, which represents the importance for prediction, and shown in the upper left