Fig. 12

Accuracy of CNN models of increasing number of layers on three datasets of increasing size. Nguyen2D corresponds to the original CNN model in [30], while Nguyen2D+1D corresponds to the same model with an extra 1D convolutional layer. Similarly, Zeng2 corresponds to the original model in [31] which has two 2D convolutional layers while Zeng3 and Zeng4 correspond to models with three and four 2D convolutional layers respectively. There is an inverse relationship between accuracy and data size with the largest dataset (motif discovery) having the lowest accuracy overall. Adding one layer with different dimension (CNN-Nguyen2D+1D) improves the accuracy slightly, but more layers of the same dimension (CNN-Zeng3 with 3 layers and CNN-Zeng4 with 4 layers) only increase the overfitting