Table 4. Mean accuracies obtained by all classifiers (apart from the regularized discriminant function analysis) using a one-vs-one and a one-vs-all multi-class approach on grey matter VBM images.

Lower and upper limits for the 95% bootstrap confidence intervals are also reported. 0.333 is the expected accuracy when no real predictive power is present.

	Algorithm	One-versus-one			One-versus-all
		Mean accuracy	5%limit	95%limit	Mean accuracy	5%limit	95%limit
Overall	aridge	0.522	0.479	0.562	0.514	0.475	0.551
	blasso	0.509	0.462	0.554	0.527	0.483	0.566
	celastic	0.517	0.478	0.557	0.527	0.493	0.559
	dL0 norm	0.527	0.464	0.588	0.566	0.525	0.602
	eSVC	0.509	0.481	0.537	0.491	0.443	0.538
	fGPC	0.501	0.452	0.547	0.515	0.463	0.563
	gRF	0.496	0.452	0.538	0.483	0.436	0.525
Controls	ridge	0.592	0.512	0.657	0.632	0.562	0.694
	lasso	0.613	0.514	0.698	0.584	0.507	0.656
	elastic	0.571	0.494	0.647	0.639	0.558	0.707
	L0 norm	0.607	0.535	0.688	0.632	0.56	0.709
	SVC	0.531	0.463	0.608	0.66	0.537	0.772
	GPC	0.633	0.543	0.72	0.656	0.564	0.741
	RF	0.621	0.548	0.688	0.68	0.591	0.764
Schizophrenia	ridge	0.635	0.544	0.738	0.631	0.544	0.74
	lasso	0.481	0.409	0.556	0.588	0.524	0.654
	elastic	0.614	0.542	0.694	0.621	0.545	0.704
	L0 norm	0.547	0.451	0.648	0.636	0.555	0.715
	SVC	0.551	0.48	0.617	0.629	0.531	0.723
	GPC	0.607	0.55	0.668	0.66	0.589	0.735
	RF	0.602	0.539	0.666	0.631	0.564	0.696
Bipolar dis.	ridge	0.372	0.285	0.448	0.282	0.185	0.363
	lasso	0.415	0.301	0.516	0.423	0.316	0.516
	elastic	0.391	0.315	0.475	0.319	0.254	0.387
	L0 norm	0.414	0.336	0.492	0.428	0.372	0.492
	SVC	0.426	0.369	0.487	0.204	0.128	0.286
	GPC	0.287	0.236	0.344	0.255	0.209	0.303
	RF	0.291	0.218	0.367	0.164	0.103	0.221

^aridge: Ridge regression,

^blasso: Lasso regression,

^celastic: Elastic net regularization,

^dL0-norm: L0-norm regularization,

^eSVC: Support vector classifier,

^fGPC: Gaussian process classifier,

^gRF: Random forest.