Support Vector Machine (SVM and SVR)

Support vector machines (SVM) are supervised learning models that find the boundary maximizing the margin between classes, and support vector regression (SVR) adapts the idea to predict continuous outcomes within an error tolerance.

서포트 벡터 머신(SVM)은 범주 간 마진을 최대화하는 경계를 찾는 지도학습 모형이며, 서포트 벡터 회귀(SVR)는 이 개념을 오차 허용 범위 안에서 연속형 결과를 예측하도록 적용한 것이다.

How do I classify or predict an outcome when the boundary between groups is complex and the sample is moderate?

집단 간 경계가 복잡하고 표본이 중간 규모일 때 결과를 어떻게 분류하거나 예측하나요?

A support vector machine separates classes with the maximum-margin hyperplane, and kernel functions such as the radial basis function allow nonlinear boundaries. Support vector regression applies the same approach to continuous outcomes, and both fit small-to-moderate samples with many features when inputs are scaled.

When to use Support Vector Machine (SVM and SVR)

• Binary or multiclass classification with complex boundaries
• Continuous prediction with SVR when robustness to small errors is desired
• High-dimensional data such as text features with moderate sample sizes

When to use something else

• Very large datasets where training time matters: use gradient boosting or linear models
• Calibrated probabilities and interpretable coefficients are needed: use logistic regression

Assumptions and requirements

• Features standardized or scaled
• Kernel and hyperparameters (C, gamma, epsilon) tuned by cross-validation
• Class imbalance handled with class weights
• Labeled training data

What the analysis reports

• Cross-validated accuracy, F1 or AUC for classification
• RMSE or MAE for regression
• Selected kernel and hyperparameters
• Number of support vectors
• Confusion matrix

How to report it

The SVM with an RBF kernel (C = [value], γ = [value]) achieved a cross-validated accuracy of [value] (SD [value]).

Related methods

• Binary Logistic Regression
• Random Forest
• K-Nearest Neighbors

Run Support Vector Machine (SVM and SVR) on Quorant