MLA-C01 Question 41
Select 3You are building a machine learning model to predict customer churn using data stored in an Amazon S3 bucket. The dataset contains categorical columns such as 'Customer_Type' and 'Region', as well as numerical columns like 'Monthly_Spend' and 'Tenure'. To prepare the data for training, you decide to perform feature engineering. Which of the following steps should you take to ensure the data is properly transformed for a machine learning model?
- A
One-hot encode the 'Customer_Type' and 'Region' columns to convert categorical data into numerical format.
- B
Normalize the 'Monthly_Spend' and 'Tenure' columns to scale them to a uniform range.
- C
Remove all rows containing null values to ensure the dataset has no missing data.
- D
Use Principal Component Analysis (PCA) to reduce the dimensionality of categorical columns.
- E
Bucketize the 'Tenure' column into ranges (e.g., 0-12 months, 13-24 months) to create new categorical features.
Show answer and explanation
Correct answers: A, B, E
Explanation
When performing feature engineering for machine learning models, it is crucial to properly handle categorical and numerical features. One-hot encoding is used to convert categorical data into numerical format. Normalizing numerical features ensures fair treatment of all features during training. Additionally, bucketizing numerical features can create new categorical features that may add value to the model. However, completely removing rows with missing values may lead to significant data loss, and PCA is not suitable for transforming categorical columns.
- A. Correct.
One-hot encoding is a standard technique to transform categorical variables into numerical format so they can be used by machine learning algorithms.
- B. Correct.
Normalizing numerical features ensures that all features contribute equally to the model and prevents features with larger ranges from dominating the training process.
- C. Incorrect.
Removing all rows with null values is not always recommended. Instead, handling missing values using techniques like mean imputation or predictive imputation is typically better.
- D. Incorrect.
PCA is generally used for dimensionality reduction of numerical features, not categorical ones. Categorical data should be encoded, not reduced using PCA.
- E. Correct.
Bucketizing numerical features like 'Tenure' can help create meaningful categories that may improve model interpretability and performance.