Bayesian parametric models for survival prediction in medical applications

Table 1 Hyperparamters that were optimized during the training process of the different models in evaluation

Model	Hyperparameter	Description	Bounds
Cox PH	alpha	Regularization parameter	(10^–5, 0.9)
Cox PH	k	K-best features	[1, n_features]
RSF	n_estimators	Number of trees	[5,25]
	tree_depth	Max depth per tree	[3, 5, 7, 9]
	k	K-best features	[1, n_features]
DeepSurv	hidden_units	Nr of hidden units	(X, n_features)
	lr	Learning rate	(10^–5, 10^–1)
	dropout	Dropout	(0.1, 0.5)
	k	K-best features	[1, n_features]
BPS Exp	priors_sd	Standard deviation of priors	(10^–1, 10¹)
BPS Exp	k	K-best features	[1, n_features]
BPS Wb	priors_sd	Standard deviation of priors	(10^–1, 10¹)
BPS Wb	k	K-best features	[1, n_features]
BPS Wb NN	n_hidden_layers	Number of hidden layers
	priors_sd	Standard deviation of priors	(10^–1, 10¹)
	k	K-best features	[1, n_features]

CoxPH Cox Proportional Hazards, RSF Random Survival Forest, BPS Bayesian Parametric Survival model, Exp Exponential, Wb Weibull, NN Neural Network

ISSN: 1471-2288