Feature Engineering Flashcards
7 cards from real MS-DS Master of Data science practice questions. Tap to flip, then mark Knew It or Still Learning — missed cards come back until you master them.
Read the first 7 Feature Engineering flashcards as text
Which technique addresses class imbalance at the feature level by creating synthetic minority samples?
Answer: SMOTE (Synthetic Minority Over-sampling Technique)
SMOTE generates synthetic minority-class samples by interpolating between existing minority-class instances in feature space, improving class balance without simply duplicating samples.
What is Weight of Evidence (WoE) encoding primarily used for?
Answer: Encoding categorical features for binary classification by measuring the strength of each category in predicting the target
WoE measures how well each category of a variable separates the binary target classes (events vs. non-events), and is widely used in credit scoring and logistic regression pipelines.
What is the purpose of the Variance Inflation Factor (VIF) in feature engineering?
Answer: Detecting multicollinearity among predictor features
VIF quantifies how much the variance of a regression coefficient is inflated due to correlation with other predictors; a high VIF (>5 or >10) indicates multicollinearity that may warrant removing or combining features.
In natural language processing feature engineering, what does TF-IDF measure?
Answer: The importance of a term in a document relative to how common it is across the entire corpus
TF-IDF (Term Frequency–Inverse Document Frequency) weighs a term's frequency in a specific document against how commonly it appears across all documents, boosting rare but informative terms.
Which feature engineering approach is most appropriate for handling a date column (e.g., '2024-03-15') in a structured dataset?
Answer: Extracting sub-features such as year, month, day of week, and is_weekend
Decomposing a date into components like year, month, day of week, and is_weekend exposes meaningful temporal patterns that a model can use, whereas a raw timestamp or label-encoded date string carries little interpretable signal.
What does a Box-Cox transformation accomplish in feature engineering?
Answer: It maps a positive-valued feature to a more normal distribution by finding an optimal power parameter λ
Box-Cox finds the optimal power λ that makes a positive feature's distribution as close to normal as possible, stabilizing variance and improving model assumptions.
Which of the following best describes 'feature hashing' (hashing trick)?
Answer: Mapping high-cardinality or text features into a fixed-size numerical vector using a hash function
Feature hashing uses a hash function to project features (such as words or categorical values) into a fixed-dimensional vector, enabling efficient handling of high-cardinality features without maintaining a vocabulary.