Multiclass Sparse Bayesian Regression for fMRI-Based Prediction

Table 3

Intersubject analysis. Explained variance 𝜁 for the different methods. The 𝑃-values are computed using a paired 𝑡-test. MCBR yields highest prediction accuracy than the two other Bayesian regularizations BRR and ARD.

Methods	Mean 𝜁	Std 𝜁	P-val/Gibbs-MCBR
SVR	0.77	0.11	.14
Elastic net	0.78	0.1	.75
BRR	0.72	0.1	.01**
ARD	0.52	0.33	.02*
Gibbs-MCBR	0.79	0.1	—
VB-MCBR	0.78	0.1	0.4

*Level of significance of the 𝑃-values.
**Level of significance of the 𝑃-values between 0.01 and 0.05.