Dynamic Principal Component Analysis with Nonoverlapping Moving Window and Its Applications to Epileptic EEG Classification

Table 5

The average classification accuracy (i.e., the proportion of all signal types correctly detected) using methods of FFPC and PCPEM with window size 𝑙=1280 for the epileptic seizure detection problem using different length of EEGs from Freiburg dataset, randomly selected from patient 1, patient 3, and both (mixtures of EEGs from both patients).

Methods	𝑟=2 hours	𝑟=3 hours	𝑟=4 hours	𝑟=5 hours
FFPC
Patient 1	1.000±0.000	0.991±0.013	0.954±0.036	0.944±0.031
Patient 3	0.996±0.009	0.992±0.014	0.984±0.027	0.981±0.026
Both	0.999±0.003	0.981±0.034	0.986±0.024	0.973±0.026
Average	𝟎.𝟗𝟗𝟖±𝟎.𝟎𝟎𝟒	𝟎.𝟗𝟖𝟖±𝟎.𝟎𝟐𝟎	𝟎.𝟗𝟕𝟓±𝟎.𝟎𝟐𝟗	𝟎.𝟗𝟔𝟔±𝟎.𝟎𝟐𝟖
PCPEM
Patient 1	0.983±0.023	0.960±0.043	0.968±0.032	0.897±0.038
Patient 3	0.980±0.020	0.977±0.036	0.955±0.048	0.934±0.055
Both	0.979±0.050	0.942±0.053	0.931±0.079	0.932±0.054
Average	𝟎.𝟗𝟖𝟎±𝟎.𝟎𝟑𝟏	𝟎.𝟗𝟔𝟎±𝟎.𝟎𝟒𝟒	𝟎.𝟗𝟓𝟏±𝟎.𝟎𝟓𝟑	𝟎.𝟗𝟐𝟏±𝟎.𝟎𝟒𝟗