Table 3.

Average accuracy of five classifiers before and after applying Box-Cox transformation using three optimization strategies for 10 times regenerated random dataset Figure 1D with different random seeds.

Classifier	Acc before [%]	Acc after [%]	Full (δ) [%]	Spherical (δ) [%]	Diagonal (δ) [%]
Linear	84.1	86.7	2.6	1.8	0.3
KNN	92.0	92.4	0.4	0.2	0.3
Bayesian	89.0	90.2	1.3	0.8	1.0
SVC	91.9	92.3	0.4	0.2	0.2
NN	92.3	92.6	0.3	0.1	0.1

Column Acc after corresponded to full optimization, which was observed as the optimal optimization. Spherical was able to get smaller but also consistent improvements. Diagonal achieved smaller gains.