MLA-C01 Question 43
Select 3You are working on a machine learning model to predict customer churn for a subscription-based service. The dataset contains customer demographic information, account usage metrics, and categorical attributes such as subscription type (e.g., 'Basic', 'Premium', 'Family'). To improve model performance, you decide to perform feature engineering. Which of the following steps are most appropriate for transforming the data to make it suitable for your model?
- A
One-hot encode the 'subscription type' feature to convert it into numerical values.
- B
Normalize numerical features such as account usage metrics to ensure they are on a similar scale.
- C
Use Label Encoding for the 'subscription type' feature to assign integer values to each category.
- D
Remove all categorical columns from the dataset since machine learning models cannot handle non-numerical data.
- E
Apply log transformation to heavily skewed numerical features to reduce the impact of outliers.
Show answer and explanation
Correct answers: A, B, E
Explanation
Feature engineering is a crucial step in preparing data for machine learning models. For categorical features without an inherent order, one-hot encoding is preferred to represent them numerically without introducing ordinal relationships. Normalizing numerical features ensures consistency in scale, which is particularly important for models sensitive to feature magnitudes, such as gradient-based algorithms. Log transformations help handle skewed numerical data by reducing outliers' influence. These steps collectively improve input data quality and model performance.
- A. Correct.
One-hot encoding is appropriate for categorical features with no ordinal relationship, such as 'subscription type', and helps in representing them numerically for machine learning models.
- B. Correct.
Normalizing numerical features ensures that features on vastly different scales do not disproportionately influence the model's performance.
- C. Incorrect.
While Label Encoding is an option, it is not ideal for non-ordinal categorical features like 'subscription type' as it may introduce unintended ordinal relationships into the model.
- D. Incorrect.
This statement is incorrect because machine learning models can handle categorical features if they are properly transformed, such as through one-hot encoding.
- E. Correct.
Log transformation is a common technique to handle skewed numerical data, reducing the impact of outliers and improving the distribution for models that assume normal distributions.