Sparse-Coding-Based Autoencoder and Its Application for Cancer Survivability Prediction

Table 6

Parameters for various training algorithms.

Algorithms	Training parameters
SAE	(i) A fully connected three-layer structure with 64 hidden neurons
SSAE	(i) A fully connected three-layer structure with 300 hidden neurons
	(ii) Aparsity parameter: 0.001
MG-RNN-AE	(i) A fully- connected three-layer structure with 10 hidden neurons
	(ii) Regularization parameters: 0.9 and 0.1
DAE	(i) A fully connected three-layer structure with 300 hidden neurons
	(ii) Regularization parameters: 0.6
RF	(i) The maximum depth per tree: 5
	(ii) The number of trees: 7
	(iii) The percentage of features used per tree:
SVM	(i) Regularization parameter: 1
	(ii) Kernel function: Radial Basis Function (RBF)
	(iii) Kernel coefficient: 0.01
	(iv) Tolerance for stopping criterion: 1e − 3
	(v) Maximum number of iterations: 500
ANN	(i) A fully connected three-layer structure with 64 hidden neurons
	(ii) The RPROP training algorithm with the maximum number of iterations 500
	(iii) The activation function is Sigmoid