SVM Parameters

C parameter:

• determines how intricate decision boundaries are.
• The Larger the number, the more intricate it is, thereby increasing the risk of ‘overfitting’ to training data
• The Smaller the number, the better it is in generalizing to test sets

Gamma parameter:

• determines how far the influence of a single training data reaches.
• With a Larger number, the training data close to decision boundary would hold more weight, thereby ignoring far-away data points in constructing decision boundary. (risk of ‘overfitting’)
• With a Smaller number, far-away data points would be considered in the construction of decision boundary. (better at generalizing to test sets)