Machine Learning Approach to Automated Quality Identification of Human Induced Pluripotent Stem Cell Colony Images

Table 2

Results of one-versus-one and one-versus-all methods when different kernels were used. True positive rates (%) are given in parentheses and accuracy (%) can be found from the last column.

Kernel/class	Bad	Good	Semigood	ACC
One-versus-all
Linear	22 (53.7%)	51 (68.9%)	18 (31.0%)	52.6%
Polynomial	23 (56.1%)	50 (67.6%)	26 (44.8%)	57.2%
Polynomial	18 (43.9%)	43 (58.1%)	22 (37.9%)	48.0%
Polynomial	23 (56.1%)	39 (52.7%)	20 (34.5%)	47.4%
Polynomial	25 (61.0%)	34 (45.9%)	17 (29.3%)	43.9%
Polynomial	23 (56.1%)	33 (44.6%)	18 (31.0%)	42.8%
RBF	25 (61.0%)	50 (67.6%)	29 (50.0%)	60.1%
One-versus-one
Linear	28 (68.3%)	52 (70.3%)	25 (43.1%)	60.7%
Polynomial	21 (51.2%)	46 (62.2%)	15 (25.9%)	47.4%
Polynomial	20 (48.8%)	45 (60.8%)	24 (41.4%)	51.4%
Polynomial	14 (34.1%)	41 (55.4%)	21 (36.2%)	43.9%
Polynomial	24 (58.5%)	32 (43.2%)	21 (36.2%)	44.5%
Polynomial	19 (46.3%)	39 (52.7%)	17 (29.3%)	43.4%
RBF	27 (65.9%)	50 (67.6%)	24 (41.4%)	58.4%