svm-platt-scaling-probability-calibration

IN premise — entries/2026/06/21/wiki-Support_vector_machine-chunk-5.md

Created 2026-06-21T09:55:55+00:00

Platt scaling is used to calibrate SVM outputs into probabilities, which is important for one-versus-all multiclass SVM.

Dependents

These beliefs depend on this one:

• OUT svm-general-purpose-classifier — SVMs serve as a general-purpose classifier with kernel nonlinearity, maximum-margin guarantees, and calibrated probability outputs via Platt scaling.
• IN svm-multiclass-requires-architectural-extension — SVMs' binary-native design requires substantial architectural extension for multiclass problems — decomposition into one-vs-all or one-vs-one subproblems, Platt scaling for probability calibration in OVA, or the unified Crammer-Singer formulation — with OVO generally outperforming OVA despite training more classifiers.