You work as a machine learning specialist at a retail shoe manufacturer. Your marketing department wants to do a promotion for a new running shoe they are about to release into their product pipeline. They need a model to predict sales of the new shoe using the purchase history of their registered customers based on past releases of new shoes.
You have decided to use a linear regression algorithm for your model. Your data has thousands of observations and 35 numeric features. While doing analysis to understand your data better, you find 25 observations that have what looks like outlier data points. After speaking to your marketing department, you learn that these values are valid. You also find several hundred observations that have some blank feature values.
How should you correct the outlier and blank feature problems?

Answer options:

A.Remove the observations with the outlier data points and replace the blank values with the null value.
B.Remove the outlier and blank value observations.
C.Remove the observations with the outlier data points and replace the blank values with the mean value.
D.Remove the observations with the outlier data points and replace the blank values with the value 0.

Answer: C
Option A is incorrect. Null values in observation should be replaced since linear regression calculations will have a problem with null values. Therefore, you would not replace empty fields with null.
Option B is incorrect. Removing the observations with blank values will reduce the accuracy of your model’s predictions since you have removed many features from the training dataset.
Option C is correct. You should remove the outlier observations. You should also replace the blank values with a meaningful value. The mean value is the best option of those listed.
Option D is incorrect.You should remove the outlier observations. You should also replace the blank values with a meaningful value. The 0 value is not the best option of those listed because the mean is invariably a better approximation than 0 for a continuous numeric value.
Reference:
Please see the Amazon Machine Learning developer guide titled Feature Processing.