Question

1 (10 points) Multiple Choice 1. Which of the following statement(s) is/are true regarding the SVM classifier? A. The margin definition in the SVM formulation can be considered as a r larization term to prevent overfitting B. Any function can be used a kernel function. C. Using a valid kernel, an SVM classifier can be trained without knowing the feature values for each sample D. The so-called "support vectors" refer to the positive and negative planes egu- 2. Which of the following statement (s) is/are true regarding the decision tree classifier? A. When training a decision tree classifier, the depth of the tree goes linearly with respect to the number of training samples B. The training objective function for a decision tree classifier has in general no analytic form to optimize for C. Tree pruning can be used to prevent overfitting. D. In general, the deeper a decision tree is, the more complex the decision boundary is

Answer 1

1)

The options A and C are true .

As in option A we can use the margin as a regularizatiom term to prevent overfitting.For regularizatiom we have to choose the parameter C correctly in order to avoid overfitting or underfitting of the model. There is a tradeoff between margin and regularization parameter C low c means high margin and high c means low margin.

Option C is correct as Svm classifier can be trained without knowing the feature value for each sample using the valid kernal. As we can use the valide kernal like poly , sigmoid, RBF kernal to transform the nonlinear feature space into linear feature space in order to get the correct feature map which can be used by the SVM for classification without knowing the actual feature value for each sample data.

Q2)

Option A and option C are correct

Option A is true as the there is a linear relationship betweem the training samples and the depth of decison tree..

Option C is correct as in case of decision trees we are using the tree pruning in order to avoid the overfitting of model..Tree pruning leads to the reduction in the complexity of final classifier and increases the accuracy by reduction of overfitting