Table 1.

Results of test set prediction using the proposed model (BHFPS) compared with 5 other methods. Two methods of basis expansions are used: cosine basis expansion and functional principal components.

	Using Cosine basis expansion				Using FPC
Method	AUC	MisR	Sens	Spec	AUC	MisR	Sens	Spec
BHFPS	0.824	26.7%	81.1%	72.6%	0.826	26.9%	80.0%	72.5%
BHVS	0.808	25.4%	74.7%	74.6%	0.814	23.7%	74.7%	76.5%
BVS	0.802	28.1%	76.8%	71.4%	0.819	30.5%	84.2%	68.0%
KNN	0.697	27.7%	62.1%	73.3%	0.718	32.1%	71.8%	74.7%
LDA	0.796	27.3%	74.7%	72.5%	0.804	25.0%	75.8%	74.9%
SVM	0.657	56.6%	85.3%	39.2%	0.679	38.4%	68.4%	61.0%

AUC: Area under ROC curve; MisR: misclassification rate; Sens: sensitivity; Speci: specificity; BHFPS: the proposed Bayesian hierarchical functional predictor selection model; BHVS: Bayesian hierarchical variable selection model; BVS: regular Bayesian variable selection model; KNN: K-nearest neighbor; LDA: linear discriminant analysis; SVM: linear support vector machine. See text for explanation of BVS and BHVS models