Multi-model inference using mixed effects from a linear regression based genetic algorithm

BMC Bioinformatics

Table 8 R ² performance on test set 2

Variable selection	LASSO			GA-OLS						GA-MM
Variable estimation	Coef (shrinkage)	OLS	MM	OLS	MM	MMI1		MMI2		OLS	MM	MMI1		MMI2
						OLS	MM	OLS	MM			OLS	MM	OLS	MM
TOP15 variables	0.667	0.734	0.712	0.707	0.707	0.708	0.708	0.709	0.710	0.709	0.702	0.705	0.696	0.706	0.698
TOP18 variables	0.690	0.731	0.713	0.721	0.718	0.716	0.714	0.722	0.719	0.768	0.770	0.742	0.742	0.747	0.750
TOP21 variables	0.742	0.760	0.765	0.736	0.730	0.722	0.717	0.732	0.726	0.777	0.775	0.746	0.744	0.751	0.752
TOP24 variables	0.745	0.771	0.768	0.732	0.728	0.720	0.716	0.727	0.723	0.762	0.761	0.743	0.740	0.748	0.749
TOP27 variables	0.767	0.788	0.788	0.721	0.725	0.720	0.717	0.732	0.726	0.770	0.768	0.744	0.741	0.758	0.755
TOP30 variables	0.777	0.789	0.787	0.768	0.772	0.731	0.729	0.747	0.743	na^c	na^c	na^c	na^c	na^c	na^c
ALL m variables ^a	0.787	0.770	0.776	na^b	na^b	0.733	0.729	0.741	0.733	na^b	na^b	0.747	0.745	0.754	0.749
	(m = 51)					(m = 193)						(m = 200)

^am is the number of variables with presence in the GA solutions or with abs(coef) > 0 in the LASSO solution path. ^bnot calculated due to singularity.
^cno model with exactly 30 mutations.
In bold the highest R² per row is indicated.

ISSN: 1471-2105