Data Preparation

QUIZ QUESTION: :

In the lessons and projects throughout this program, you've encountered various aspects of data preparation; balanced classes, randomization, train/test splits and others. Here's a quiz to review why these steps are necessary (drag and drop the colored boxes to the appropriate location in the right-hand column).

ANSWER CHOICES:

Prepare a balanced dataset, i.e., have as many positive as negative examples, or in the case of multi-class problems, roughly the same number of cases of each class.

Random Shuffling of the data

Splitting the data into a training and testing set

Normalization of features, typically to zero mean and unit variance

Data Preparation Step	Purpose
To avoid having your algorithm simply classify everything as belonging to the majority class.
To avoid problems due to ordering of the data
To avoid individual features or sets of features dominating the response of your classifier
To estimate generalization of the model to new data

SOLUTION:

Data Preparation Step	Purpose
To avoid having your algorithm simply classify everything as belonging to the majority class.	Prepare a balanced dataset, i.e., have as many positive as negative examples, or in the case of multi-class problems, roughly the same number of cases of each class.
To avoid problems due to ordering of the data	Random Shuffling of the data
To avoid individual features or sets of features dominating the response of your classifier	Normalization of features, typically to zero mean and unit variance
To estimate generalization of the model to new data	Splitting the data into a training and testing set