Hands-On Exercise: Data Preparation for ML CT-AI Practice Questions (Page 3)

Questions 11–15

1. 11expert · hard

   A data science team is building a model for a healthcare dataset. They have a 'patient_id' column, a 'diagnosis_date' column, and several clinical features. They plan to train a gradient boosting model. Which data preparation step is most critical to avoid data leakage?

   • ARemove duplicate rows from the dataset.
   • BSort the data by 'diagnosis_date' before splitting.
   • CEnsure all preprocessing steps (e.g., imputation, scaling) are fitted only on the training data and applied to validation/test data.
   • DDrop the 'patient_id' column because it is unique and not predictive.
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   What is feature engineering?

   • AThe process of splitting data into training and test sets
   • BThe process of selecting the best ML algorithm
   • CThe process of creating new features or transforming existing ones to improve model performance
   • DThe process of cleaning missing values from the dataset
   Reveal Solution Discussion Select an answer first
4. 13application · easy

   A junior data scientist is starting a new ML project with a raw dataset. They want to follow a proper data preparation workflow. What is the correct order of the main steps?

   • AFeature engineering, data cleaning, data splitting, data transformation.
   • BData cleaning, data transformation, feature engineering, data splitting.
   • CData transformation, data cleaning, feature engineering, data splitting.
   • DData splitting, data cleaning, data transformation, feature engineering.
   Reveal Solution Discussion Select an answer first
5. 14expert · hard

   An ML engineer is preparing a dataset for a neural network. The dataset has a 'country' column with 50 categories and a 'product_rating' column with values from 1 to 5 (integer). The engineer wants to use embedding layers for the country feature. Which preprocessing approach is most appropriate?

   • ALabel encode 'country' to integer indices and use an embedding layer in the network; scale 'product_rating' to [0,1].
   • BOne-hot encode 'country' and scale 'product_rating' to [0,1].
   • CLabel encode 'country' and use 'product_rating' as-is.
   • DOne-hot encode 'country' and use 'product_rating' as-is.
   Reveal Solution Discussion Select an answer first
6. 15application · medium

   A team is preparing a dataset for a credit risk model. The 'age' column has some values like -5 and 200, and the 'income' column has 10% missing values. The team decides to handle these issues. Which approach is most appropriate?

   • ARemove all rows with any outlier or missing value.
   • BReplace age outliers with the mean and impute missing income with 0.
   • CImpute missing income with the mean and leave age outliers as-is.
   • DReplace age values outside a plausible range (e.g., 18-100) with the median, and impute missing income with the median.
   Reveal Solution Discussion Select an answer first