Data Preparation & Preprocessing Exam Questions with Correct Answers Latest Update 2025-
2026
Tidy data - Answers Each row is an observation, each column is a variable or dimension, each
cell is a piece of data or value.
Train-test split - Answers To see how the model generalizes to new/unseen data.
Stratification - Answers Contains the same proportion of values in each class in the training and
test.
Undersampling - Answers Under sample from the larger class until the two proportions are
equal.
Oversampling - Answers Over sample from the smaller class; balancing the data out a bit more.
SMOTE - Answers Synthetic minority oversampling technique: generates or synthesizes new
examples from existing data.
Data leakage - Answers Some training data ends up in testing data.
Train test contamination - Answers Oversampling/SMOTE before train test split.
Target leakage - Answers Inclusion of future info or inappropriate feature selection.
Data quality types - Answers Consistency, missingness, uniqueness, duplicates, integrity.
Types of missingness - Answers MCAR: missing completely at random; MAR: missing at
random; MNAR: missing not at random.
Data imputation - Answers Solution to missingness; fill with median or mean.
MICE - Answers Multivariate imputation by chained equations.
Types of skewness - Answers Normal, left (tail to the left), right (tail to the right), bimodal, flat,
outliers.
Bimodal data - Answers Two peaks.
Ordinal encoding - Answers Encode each categorical feature as an int, implying a ranking or
order to data that has none.
One-hot encoding - Answers Each category becomes a feature (column); 1 means it is in that
category, 0 means it is not.
Count vectorization - Answers One column for every word in your vocab; counts the words in a
document.
2026
Tidy data - Answers Each row is an observation, each column is a variable or dimension, each
cell is a piece of data or value.
Train-test split - Answers To see how the model generalizes to new/unseen data.
Stratification - Answers Contains the same proportion of values in each class in the training and
test.
Undersampling - Answers Under sample from the larger class until the two proportions are
equal.
Oversampling - Answers Over sample from the smaller class; balancing the data out a bit more.
SMOTE - Answers Synthetic minority oversampling technique: generates or synthesizes new
examples from existing data.
Data leakage - Answers Some training data ends up in testing data.
Train test contamination - Answers Oversampling/SMOTE before train test split.
Target leakage - Answers Inclusion of future info or inappropriate feature selection.
Data quality types - Answers Consistency, missingness, uniqueness, duplicates, integrity.
Types of missingness - Answers MCAR: missing completely at random; MAR: missing at
random; MNAR: missing not at random.
Data imputation - Answers Solution to missingness; fill with median or mean.
MICE - Answers Multivariate imputation by chained equations.
Types of skewness - Answers Normal, left (tail to the left), right (tail to the right), bimodal, flat,
outliers.
Bimodal data - Answers Two peaks.
Ordinal encoding - Answers Encode each categorical feature as an int, implying a ranking or
order to data that has none.
One-hot encoding - Answers Each category becomes a feature (column); 1 means it is in that
category, 0 means it is not.
Count vectorization - Answers One column for every word in your vocab; counts the words in a
document.
