Supervised Machine Learning Based Noninvasive Prediction of Atrial Flutter Mechanism from P-to-P Interval Variability under Imbalanced Dataset Conditions

Table 7

Feature selection methods (both filter and wrapper) at the original dataset.

	Original dataset
	Filter method	Wrapper method
Features	p value	Feature score
		LDA	LOG	SVM
Mean	0.80	0.9	1	1
Median	1	1	1	1
Mode	0.97	0.9	1	1
Std	0.57	0.9	1	1
Var	0.57	0.9	1	1
Skew	0.34	0.9	0.9	1
Kurt	0.62	0.8	0.7	1
PDmin	0.39	1	1	1
PDmax	0.69	0.7	0.8	1
Interval	0.28	0.8	0.9	1

Note. In the filter method, we used the Wilcoxon rank-sum test.