Applications of Bayesian Gene Selection and Classification with Mixtures of Generalized Singular 𝑔-Priors

Table 2

Performance comparison of different procedures with LOOCV for the colon cancer study.

Methods	No. of genes	LOOCV error rate	LOOCV accuracy
Bayesian 𝑔-prior	6	0.1452 (9/62)	0.8548 (53/62)
Bayesian 𝑔-prior	10	0.1452 (9/62)	0.8548 (53/62)
Bayesian 𝑔-prior	14	0.1129 (7/62)	0.8871 (55/62)
SVM^a	1000	0.0968 (6/62)	0.9032 (56/62)
Classification tree^b	200	0.1452 (9/62)	0.8548 (53/62)
1-Nearest-neighbor^b	25	0.1452 (9/62)	0.8548 (53/62)
LogitBoost, estimated^b	25	0.1935 (12/62)	0.8065 (50/62)
LogitBoost, 100 iterations^b	10	0.1452 (9/62)	0.8548 (53/62)
AdaBoost, 100 iterations^b	10	0.1613 (10/62)	0.8387 (52/62)
MAVE-LD^c	50	0.1613 (10/62)	0.8387 (52/62)
IRWPLS^d	20	0.1129 (7/62)	0.8871 (55/62)
SGLasso^e	19	0.1290 (8/62)	0.8710 (54/62)
MRMS + SVM + D1^f	5	0.1290 (8/62)	0.8710 (54/62)
MRMS + SVM + D2^f	33	0.1452 (9/62)	0.8548 (53/62)
𝑡-test + probit regression	6	0.1452 (9/62)	0.8548 (53/62)
𝑡-test + probit regression	10	0.1774 (11/62)	0.8226 (51/62)
𝑡-test + probit regression	14	0.2258 (14/62)	0.7742 (48/62)

aProposed by Furey et al. [41].
^bProposed by Dettling andBühlmann [42].
^cProposed by Antoniadis et al. [43].
^dProposed by Ding and Gentleman [44].
^eProposed by Ma et al. [45].
^fProposed by Maji and Paul [46].