AWS Certified Machine Learning – Specialty — Question 272

A medical device company is building a machine learning (ML) model to predict the likelihood of device recall based on customer data that the company collects from a plain text survey. One of the survey questions asks which medications the customer is taking. The data for this field contains the names of medications that customers enter manually. Customers misspell some of the medication names. The column that contains the medication name data gives a categorical feature with high cardinality but redundancy.

What is the MOST effective way to encode this categorical feature into a numeric feature?

Answer options

• A. Spell check the column. Use Amazon SageMaker one-hot encoding on the column to transform a categorical feature to a numerical feature.
• B. Fix the spelling in the column by using char-RNN. Use Amazon SageMaker Data Wrangler one-hot encoding to transform a categorical feature to a numerical feature.
• C. Use Amazon SageMaker Data Wrangler similarity encoding on the column to create embeddings of vectors of real numbers.
• D. Use Amazon SageMaker Data Wrangler ordinal encoding on the column to encode categories into an integer between 0 and the total number of categories in the column.

Correct answer: C

Explanation

Amazon SageMaker Data Wrangler similarity encoding is ideal for high-cardinality categorical features containing spelling variations and errors, as it converts text strings into numerical vector embeddings based on string similarity. One-hot encoding (Options A and B) would result in a massive, sparse feature space because of the high cardinality and spelling mistakes. Ordinal encoding (Option D) is unsuitable here because it implies an arbitrary ordered relationship between medications and does not resolve spelling redundancies.