Skip to main content

Table 2 Elastic Net - best results, after Soft Impute with λ=100

From: Parameters of stochastic models for electroencephalogram data as biomarkers for child’s neurodevelopment after cerebral malaria

Feature set

LOOCV MSE

Number of nonzero coefficients

Sample features with non-zero coefficients

54 anthropomentic, socio-economic, and medical non-EEG features

0.3982

15

Weight, hemoglobin level, weight, BCS, the HOME score, white blood cell count, cerebrospinal fluid levels of interleukin (IL)-1 receptor antagonist (RA), IL-6, RANTES (an acronym for Regulated on Activation, Normal T Expressed and Secreted), IL-8, and plasma levels of vascular endothelial growth factor and von Willebrand factor.

362 traditional EEG features in frequency bands

0.5285

62

Features include: fp2 in T6-CZ, FZ-CZ and P4-CZ, fp1 F4-CZ, wave energy for theta in channel T5-CZ, variance in theta for FP2-CZ, variance in alpha frequency band for T5-CZ.

150 stochastic features for increment processes in frequency bands

0.1511

85

12 coefficients have the absolute values over 0.5. Top 5 coefficients are for the channels FP2-CZ, O1-CZ, T6-CZ, FZ-CZ. Channels F8-CZ and T6-CZ appear most often among the 12 top coefficients by absolute value, 4 and 3 times respectively. Four of the top 12 are for fitting a stochastic process model for the a7 frequency band.