Question

A data science team is preparing a customer churn dataset stored in Amazon S3. Several numeric columns (such as monthly usage minutes) contain missing values for roughly 8% of records. The team computes the median of each affected column from the training data and uses it to fill the gaps. They plan to deploy the model behind a SageMaker real-time endpoint that will receive individual customer records that may also have missing values. Which approach best ensures the imputation is applied consistently during both training and inference?

Accepted Answer

Persist the median values computed from the training set and apply those same stored values in the inference preprocessing logic served with the endpoint

Answer

Recompute the median from each incoming inference request and use it to fill any missing values in that request

Answer

Drop any inference records that contain missing values so the model only scores complete records

Answer

Replace missing values with zero during inference since the model was already trained to handle numeric columns