DAA-C01 Question 107
Single answerClassificationA retail analytics team wants to predict whether a customer will respond to a new loyalty campaign. They are building the model entirely in Snowflake using Snowpark ML and need a binary classification approach that supports probability outputs for each customer so marketing can target only customers above a confidence threshold. The source table contains a mix of numeric features, one-hot encoded categorical features, and a target column named RESPONDED with values 0 or 1. Which approach best satisfies the requirement?
- A
Train a Snowpark ML LogisticRegression model using RESPONDED as the label, then use prediction probability output to score customers and filter by a threshold.
- B
Train a Snowpark ML LinearRegression model on RESPONDED, then round the predicted numeric values to 0 or 1 for campaign targeting.
- C
Use KMeans clustering to group customers into two clusters and treat one cluster as responders and the other as non-responders.
- D
Use PCA to reduce the feature set to one component and classify customers based on whether the component value is positive or negative.
Show answer and explanation
Correct answer: A
Explanation
The key requirement is supervised binary classification with probability outputs so the business can apply a confidence threshold. Logistic regression is a common and appropriate choice for this scenario because it is specifically intended for binary outcomes and produces class scores that can be used for targeted decisioning. In contrast, linear regression predicts continuous values and is not a best-practice classifier, while KMeans and PCA are unsupervised methods that do not directly solve labeled classification problems. Snowflake Snowpark ML supports classification workflows where labeled data is used to fit models and generate predictions for downstream analytics and operationalization. Best practice is to choose an algorithm aligned to the prediction objective first, then evaluate performance using classification metrics such as precision, recall, F1, or ROC-related measures rather than treating the problem as regression or clustering.
- A. Correct.
Correct. Logistic regression is a standard supervised algorithm for binary classification and is appropriate when the label is 0/1. In Snowpark ML, this aligns with a classification workflow and supports probability-style outputs that can be used for threshold-based decisioning, which is exactly what the marketing team needs.
- B. Incorrect.
Incorrect. Linear regression is designed for continuous numeric prediction, not classification. Although some users try to repurpose it by rounding outputs, that does not produce a proper classification model and does not provide well-formed class probabilities for thresholding. This is a common misconception when the target column happens to be coded as 0 and 1.
- C. Incorrect.
Incorrect. KMeans is an unsupervised clustering algorithm and does not use the known RESPONDED label during training. Even if two clusters are requested, cluster membership does not correspond reliably to responder versus non-responder classes. This confuses segmentation with supervised prediction.
- D. Incorrect.
Incorrect. PCA is a dimensionality reduction technique, not a classification algorithm. It can sometimes be used as a preprocessing step before model training, but by itself it does not learn a decision boundary from the RESPONDED label. Using the sign of a principal component is arbitrary and not a valid binary classification approach.