When should an SVM favor correct classifications over a wide boundary? The C parameter sets that trade-off: larger values emphasize classification errors; smaller values emphasize distance between the boundary lines. section 11.1
Error formula = C · (Classification Error) + (Distance Error)
The error formula weights classification error by C while leaving distance error unweighted. section 11.1
Compare two C values
On a sample dataset, C=0.01 makes some classification errors and reaches 0.867 accuracy. C=100 emphasizes classifying points correctly and has higher accuracy. section 11.2
