Figure 2
Experiment results of models. (A) k-fold cross-validation was used for setting hyper-parameters and evaluating generalizations. (B) The results from independent repeated experiments indicated that there was little prospect of prediction by using SVM with different kernel functions and RF. The mean performance of libD3C, which is the latest ensemble classifier, was slightly higher than 20%. The validation accuracy of CNN and Softmax was 40% versus 70%. (C) The learning curve of experimental results under different hyper-parameter settings, including learning rate, training threshold, validation threshold, and regularization weight. The best results were 0.06 of the learning rate, 0.30 of the training threshold, 0.06 of the validation threshold, and 1 of the regularization weight. (D) Softmax result under best hyper-parameter setting. The training accuracy was over 80%, and the validation accuracy was approximately 70%. (E) Confusion matrix representing Softmax classification performance. Each column of the matrix represents the instances in a predicted class, whereas each row represents the instances in an actual class.
