Supervised Machine Learning Based Noninvasive Prediction of Atrial Flutter Mechanism from P-to-P Interval Variability under Imbalanced Dataset Conditions
Table 5
Comparison in the percentage of the augmented dataset by descriptive statistics on the minority data (focal AFL).
Mean
Variance
Skewness
SMOTE
Modified
Smoothed-bootstrap
SMOTE
Modified
Smoothed-bootstrap
SMOTE
Modified
Smoothed-bootstrap
Difference between statistics of the original dataset and the augmented dataset (in percentage)
100
−0.19
−0.17
0.43
25.52
10.65
−4.36
18.34
19.98
3.45
200
−0.19
−0.14
0.43
25.74
9.69
−4.27
23.87
19.74
4.69
300
−0.18
−0.17
0.42
25.61
10.59
−4.01
20.70
18.82
4.43
400
−0.20
−0.16
0.41
25.93
10.59
−3.83
20.35
18.92
4.85
500
−0.19
−0.17
0.44
25.95
9.95
−4.26
21.51
20.11
4.51
600
−0.20
−0.15
0.41
25.93
10.26
−3.78
21.53
20.01
4.81
700
−0.19
−0.17
0.42
25.45
10.77
−3.97
21.00
20.14
4.42
800
−0.19
−0.17
0.41
25.85
10.55
−4.04
22.86
20.80
5.15
There are three augmentation techniques (Smote, Modified-Smote, and Smoothed-Bootstrap) for each parameter (Mean, Variance, and Skewness). The technique with the minimum difference is indicated in bold font.
