EMD on MMSE probe 2

Raven 2 class

• Dane: emd_mmse2_c2.csv
• Features: stats (avg, std, sk, kr, min, max) + stats on emd components distr (avg, std, sk, kr)
• Classes: raven scaled < 11.3 (dist: class 0 (21 vect 51.2%, 22 vect 48.8% )

Classification

• data preparation: standarization
• 10x5CV stratiffied

QPC visualizations

w =

Columns 1 through 7

 -0.1973    0.2776   -0.0025    0.0552   -0.1026   -0.1199   -0.3357
 -0.1870    0.3552   -0.1334    0.0696   -0.5460   -0.0277   -0.4003

Columns 8 through 10

 -0.1334    0.4272   -0.7369
 -0.1687   -0.5491    0.1647

After transformation in 2D space

• GM 10x5CV kNN euclid k=3 acc 78.9 % / 11.6

QPC on 5 best features

• sele cted features 2 (std), 5 (min) , 7 (sStd), 9 (sSK), 10 (sKr)

w3 =

  0.0507   -0.4182   -0.4727    0.3955   -0.6653
  0.2500   -0.5526   -0.3876   -0.6441    0.2589
  

1D projection on 4 features

• features: 5 (min), 7 (sAvg), 9 (sSk), 10 (sKr)
• projection

w = [ 0.70  0.34  -1.09  3.24 ]
y = data(:,[5 7 9 10]) * w'

• rule y > 8.13 then 1 else 0 → 72.1%
• rule y > 8.13 AND y < 8.48 then 1 else 0 → 83.7%