Skip to main content

Table 3 Individual and pairwise stability of the base FS algorithms used in the ensemble. Results are averaged over the 10 × 5 stratified CV and m subsets (for each possible subset size) for the 2-years (upper values), 3-years (middle values) and 4-years (bottom values), using ADNI data

From: Neuropsychological predictors of conversion from mild cognitive impairment to Alzheimer’s disease: a feature selection ensemble combining stability and predictability

 

ReliefF

MIM

CMIM

MRMR

Chi-Squared

SVM-RFE

LL21

ReliefF

0.784 ± 0.127

0.755 ± 0.147

0.728 ± 0.158

MIM

0.601 ± 0.192

0.863 ± 0.112

0.589 ± 0.215

0.862 ± 0.108

0.570 ± 0.208

0.861 ± 0.114

CMIM

0.589 ± 0.243

0.704 ± 0.124

0.774 ± 0.152

0.533 ± 0.242

0.667 ± 0.139

0.758 ± 0.164

0.479 ± 0.23

0.618 ± 0.168

0.739 ± 0.168

MRMR

0.446 ± 0.211

0.301 ± 0.162

0.396 ± 0.174

0.858 ± 0.054

0.371 ± 0.176

0.252 ± 0.144

0.379 ± 0.171

0.852 ± 0.064

0.367 ± 0.185

0.237 ± 0.161

0.390 ± 0.165

0.858 ± 0.054

Chi-Squared

0.583 ± 0.1850

0.646 ± 0.118

0.529 ± 0.158

0.335 ± 0.212

0.871 ± 0.136

0.591 ± 0.195

0.668 ± 0.130

0.514 ± 0.159

0.289 ± 0.193

0.874 ± 0.149

0.574 ± 0.217

0.668 ± 0.135

0.497 ± 0.167

0.286 ± 0.194

0.875 ± 0.148

SVM-RFE

0.233 ± 0.097

0.184 ± 0.089

0.226 ± 0.107

0.323 ± 0.087

0.141 ± 0.068

0.302 ± 0.142

0.219 ± 0.100

0.233 ± 0.089

0.251 ± 0.086

0.272 ± 0.064

0.184 ± 0.07

0.269 ± 0.076

0.217 ± 0.089

0.201 ± 0.089

0.227 ± 0.098

0.277 ± 0.059

0.193 ± 0.084

0.273 ± 0.097

LL21

0.618 ± 0.224

0.617 ± 0.215

0.552 ± 0.236

0.414 ± 0.215

0.584 ± 0.202

0.200 ± 0.088

0.908 ± 0.056

0.606 ± 0.22

0.613 ± 0.229

0.539 ± 0.241

0.353 ± 0.185

0.574 ± 0.203

0.206 ± 0.099

0.887 ± 0.064

0.546 ± 0.253

0.565 ± 0.242

0.475 ± 0.235

0.324 ± 0.186

0.541 ± 0.209

0.184 ± 0.084

0.856 ± 0.087