Fig. 3. Classification problems: Comparing approximation and classification performances of SPA (blue curves) to the common methods on biomedical applications (40, 41).

Common methods include K-means clustering (dotted lines), SOM (brown), pattern recognition NNs (dashed lines), GMMs (cyan), density-based clustering (gray dotted lines with crosses), and Bayesian models (Eq. 1) (Bayes; dotted lines). Approximation error is measured as the multiply cross-validated average squared Euclidean norm of difference between the true and the discretized representations for validation data. Classification error is measured as the multiply cross-validated average total variation norm between the true and the predicted classifications for validation data. WDBC, Wisconsin Diagnostic Breast Cancer database.