Databricks Machine Learning Associate Question 541
Select 3You are tasked with developing a machine learning model to predict customer churn using Databricks. During the model development process, you notice that your dataset contains several categorical features. Which of the following steps are appropriate to handle these categorical features before training the model?
- A
Use one-hot encoding to convert the categorical variables into binary columns.
- B
Drop all categorical features from the dataset to simplify the model.
- C
Use label encoding to assign unique numerical values to each category.
- D
Apply normalization directly to the categorical features to scale their values.
- E
Use feature hashing to convert categorical features into fixed-length numerical representations.
Show answer and explanation
Correct answers: A, C, E
Explanation
Handling categorical features is a crucial step in model development. One-hot encoding, label encoding, and feature hashing are all appropriate techniques to preprocess categorical data, depending on the specific use case and dataset characteristics. Dropping features or applying normalization directly to categorical data are not recommended actions as they either lose information or fail to treat the data meaningfully.
- A. Correct.
One-hot encoding is a common and effective method to handle categorical features by creating binary columns for each category. This ensures the model can interpret the categorical data correctly.
- B. Incorrect.
Dropping categorical features might result in the loss of valuable information, which could negatively impact the model's performance. This is not a recommended approach unless the features are irrelevant.
- C. Correct.
Label encoding is a valid approach as it assigns a unique numerical value to each category. However, it may introduce ordinal relationships that some models might misinterpret.
- D. Incorrect.
Normalization is not applicable to categorical features because they are not numerical in nature, and scaling their values directly would not add meaningful information for the model.
- E. Correct.
Feature hashing is another valid technique to handle categorical features, especially when dealing with high-cardinality data. It converts categorical values into fixed-length numerical representations efficiently.