Support Vector Regression (SVR)

Finds a function that deviates from actual values by at most epsilon, while being as flat as possible.

When to use:

• Small to medium datasets
• High-dimensional data (many features)
• Complex non-linear patterns
• Want robust predictions

Strengths: Effective in high dimensions, memory efficient, versatile kernels for non-linearity Weaknesses: Slow on large datasets, sensitive to feature scaling, many hyperparameters

Model Parameters

C (default: 1.0) Regularization parameter. Controls trade-off between smooth function and training accuracy.

• Low (0.1-1): Smooth function, may underfit
• Default (1.0): Balanced
• High (10-100): Fits training data closely, may overfit

Kernel Transformation function:

• rbf: Radial basis function (default, handles non-linear)
• linear: For linear relationships (faster)
• poly: Polynomial kernel (flexible but slow)

Gamma Kernel coefficient:

• scale: 1 / (n_features * X.var()) (default)
• auto: 1 / n_features
• Custom value: Higher = more complex

Epsilon (default: 0.1) Width of the tube where no penalty is given. Larger = more errors tolerated.