Table 14 Comparison of our results with those reported in the literature for the MCI versus HC classification.
References | Feature extraction methods | Classification methods | Data used | No. of channels | CA (%) |
|---|---|---|---|---|---|
Power, relative power, power ratio for different bands | Neurofuzzy + KNN | 11 MCI/16 HC | 3 | 88.89% using hold-out validation | |
DWT | Decision Tree (C4.5) | Own data,37MCI/ 23 HC | 19 | 93.3% using tenfold CV83.3% using hold out | |
Supervised dictionary learning with spectral features, named CLC-KSVD | Same in24 | 3 | 88.9% using hold-out validation | ||
Power spectral features | KNN | Same in24 | 19 | 81.5% | |
SWT + statistical features | SVM | Data from24, 11 MCI/ 11 HC | 19 | 96.94% based on intra-subject validation | |
Permutation entropy and auto-regressive | ELM | Same in24 | 19 | 98.78% using tenfold CV | |
kernel Eigen-relative-power | SVM | 24 MCI/ 27 HC | 5 | 90.2% using LOSO CV | |
DWT + PSD + coherence | Bagged Trees | Same in24 | 19 | 96.5% using fivefold CV | |
Power intensity for each high and low-frequency band | KNN | Same in36 | 19 | 95.0% using tenfold CV | |
– | LSTM | Same in24 | 19 | 96.41% using fivefold CV | |
Several features using 10 measures | SVM | Private data,21 MCI/ 21 HC | 8 | 86.85% using LPSO CV | |
Spectral, functional connectivity, and nonlinear features | SVM | 18 MCI/ 16 HC | 19 | 99.4% using tenfold CV | |
DWT leader | AdaBoostM1 | Same in32 | 19 | 93.50% using tenfold CV | |
EMD + Log energy entropy | KNN | 19 | 97.60% using tenfold CV | ||
– | CNN | Same in24 | 19 | 84.28% using LOSO CV | |
Present study | VMD + TeEng | SVM | Data from24, 11 MCI/ 13 HC | 7 | 95.28% and 95.83%, respectively, using LOSO CV and NSGA-II |
DWT + TeEng | 8 |
