Exam AWS Certified Machine Learning - Specialty topic 1 question 236 discussion

C. Quantile binning: Quantile binning (or discretization) involves dividing a continuous variable into bins based on quantiles. This can be useful for handling skewed data by distributing the data more evenly across the bins. However, this method transforms the numerical feature into a categorical one, which might not be ideal for preserving the ordinal nature and the detailed variance of the 'duration' feature in a regression model. If the choice must be made from the given options, Option C (Quantile binning) might be the most suitable, albeit not ideal, as it can at least help in dealing with skewed distributions by distributing the data across bins more evenly. However, the data scientist should consider logarithmic or polynomial transformations for a more direct approach to addressing non-linearity.

https://docs.aws.amazon.com/machine-learning/latest/dg/data-transformations-reference.html

A. NO - One-hot encoding is for featurization of categories B. NO - C. YES - Quantile binning can make data linear (https://docs.aws.amazon.com/machine-learning/latest/dg/data-transformations-reference.html#quantile-binning-transformation) D. NO - Normalization will recenter the data, not change the relationship

quantile binning

C is correct

C is the best answer I guess

the correct answer is C, Quantile binning. This transformation divides the data into quantiles (equal-sized intervals) based on the values of the feature (in this case, duration) and replaces the values with the bin number. This transformation can help capture non-linear relationships between features by creating more representative categories for skewed data. The transformed data can then be used to train a non-linear regression model, such as a polynomial regression, to better predict future book sales.