4. [10 points] On classification, circle T (true) or F (false) for each statement. Note: Bonus...

Transcribed Image Text:

4. [10 points] On classification, circle "T" (true) or "F" (false) for each statement. Note: Bonus points can be given if false statements are corrected. T/F: Logistic regression works well for perfectly separable data, that is, a hyperplane in the feature space can separate the two classes in the training set. Note: hyperplane is a line in case of a 2D region, a plane in case of a 3D space, a 3 dimensional shape such as a sphere in case of 4D space and so on. T/F: The decision boundary using both KNN method (K nearest neighbors) and LDA (linear discriminant analysis) method is linear, that is, a hyperplane of the feature space. T/F: KNN method often performs much better for data with large p than small Р, where p is the number of features. T/F: Sensitivity of a classifier in a binary classification problem (y = 0 or 1) is the proportion of observations in class 0 that are correctly classified to be 0 by that classifier. T/F: Similar as principal component regression for continuous y, we can use principal component analysis to construct a small number of principal components, that is, derived features, and use the new features for classification. The method is called principal component classification. 4. [10 points] On classification, circle "T" (true) or "F" (false) for each statement. Note: Bonus points can be given if false statements are corrected. T/F: Logistic regression works well for perfectly separable data, that is, a hyperplane in the feature space can separate the two classes in the training set. Note: hyperplane is a line in case of a 2D region, a plane in case of a 3D space, a 3 dimensional shape such as a sphere in case of 4D space and so on. T/F: The decision boundary using both KNN method (K nearest neighbors) and LDA (linear discriminant analysis) method is linear, that is, a hyperplane of the feature space. T/F: KNN method often performs much better for data with large p than small Р, where p is the number of features. T/F: Sensitivity of a classifier in a binary classification problem (y = 0 or 1) is the proportion of observations in class 0 that are correctly classified to be 0 by that classifier. T/F: Similar as principal component regression for continuous y, we can use principal component analysis to construct a small number of principal components, that is, derived features, and use the new features for classification. The method is called principal component classification.

Answer rating: 100% (QA)

1 True Logistic regression has traditionally been used to come up with a hyperplane tha... View the full answer