Markup | Genome Research

Table 4.

Accuracy of classification methods

Model	Criterion	All		EUR		EAS		AMR		SAS		AFR		OCE		WAS
Model	Criterion	TR	TS	TR	TS	TR	TS	TR	TS	TR	TS	TR	TS	TR	TS	TR	TS
PCA	${\arg \min}_{k} ‖ z_{n} - c_{k} ‖_{2}^{2}$	78.6	74.1	64.9	66.3	71.1	74.3	87.2	77.8	58.5	57.2	97.5	93.4	95.4	76.6	75.6	73.4
VAE		93.2	85.7	81.7	78.6	96.9	96.3	99.5	92.1	81.6	78.5	99.3	96.8	99.4	71.7	94.0	86.1
C-VAE	${\arg \min}_{k} ‖ x_{n} - {\hat{x}}_{n}^{(k)} ‖_{1}$	93.4	78.0	84.4	70.6	96.4	92.1	99.9	92.4	84.4	76.4	99.9	96.8	100	62.8	88.5	54.7
C-VAE	arg max_k p(Y = k\|x_n, θ)	97.5	87.1	96.4	87.1	98.5	95.5	100	97.4	90.0	81.2	99.4	94.9	100	79.8	98.1	73.5
Y-VAE	${\arg \min}_{k} ‖ x_{n} - {\hat{x}}_{n}^{(k)} ‖_{1}$	98.9	83.2	96.9	81.5	99.6	96.2	99.9	87.2	98.5	90.1	100	98.4	100	68.6	97.5	59.9
Y-VAE	arg max_k p(Y = k\|x_n, θ)	99.1	85.2	97.6	84.3	99.7	96.6	100	90.9	98.2	88.5	100	98.2	100	72.7	98.3	65.0

[i] TR refers to accuracy computed on training data and TS on test data, accordingly. The values represent the accuracy in %. Note that regular VAE, C-VAE, and Y-VAE have 10,371,760, 10,378,928, and 72,602,320 parameters, respectively. Bold values indicate the highest accuracy (best performance) on test samples across the compared models and criteria.