Support Vector Regression

Support Vector Regression (SVR) fits a function within an epsilon-tube around the data, tolerating small errors while penalizing larger deviations. Kernel functions allow it to model complex non-linear relationships.

SVR works best on small to medium-sized datasets where high accuracy is needed.

Choice of kernel is critical:

• Linear kernel → Behaves like linear regression
• RBF kernel → Captures non-linear patterns
• Poly kernel → Models curved relationships

Use Cases

• Energy prediction models
• Performance optimization problems with complex geometry interactions
• Façade performance prediction based on multiple interacting parameters