AML AML Feature Engineering & Data Preprocessing 2 — Questions and Answers
Question 1: What is the curse of dimensionality in machine learning?
- The difficulty of visualizing high-dimensional data
- The phenomenon where data becomes sparse as dimensions increase, degrading model performance (Correct answer)
- The computational cost of training deep neural networks
- The problem of having too many training samples
Correct answer: The phenomenon where data becomes sparse as dimensions increase, degrading model performance
As feature dimensions increase, data points become increasingly sparse, making distance-based algorithms less effective.
Question 2: Which encoding technique assigns integer values to categories that have a natural order?
- One-hot encoding
- Binary encoding
- Ordinal encoding (Correct answer)
- Frequency encoding
Correct answer: Ordinal encoding
Ordinal encoding maps ordered categories such as low/medium/high to integers that preserve their natural rank.
Question 3: What is the purpose of a train-validation-test split in model development?
- To increase the size of the training set
- To tune hyperparameters and evaluate final model performance on unseen data (Correct answer)
- To perform feature engineering
- To balance class distributions
Correct answer: To tune hyperparameters and evaluate final model performance on unseen data
The validation set enables hyperparameter tuning while the test set provides an unbiased final performance estimate.
Question 4: Which technique generates synthetic minority class samples to address class imbalance?
- Random undersampling
- SMOTE (Correct answer)
- Stratified sampling
- Bootstrapping
Correct answer: SMOTE
SMOTE creates new synthetic samples by interpolating between existing minority class examples rather than duplicating them.
Question 5: What does a log transformation primarily help with during feature engineering?
- Encoding categorical variables
- Reducing the skewness of right-skewed distributions (Correct answer)
- Normalizing binary features
- Handling missing data
Correct answer: Reducing the skewness of right-skewed distributions
Log transformation compresses large values and expands small ones, making right-skewed distributions more symmetric.
Question 6: What is a machine learning data pipeline?
- A visualization tool for feature distributions
- An automated sequence of data processing steps from ingestion to model-ready format (Correct answer)
- A method for hyperparameter tuning
- A storage system for training datasets
Correct answer: An automated sequence of data processing steps from ingestion to model-ready format
A data pipeline automates and chains data collection, cleaning, transformation, and feature engineering steps for reproducibility.
AML AML Feature Engineering & Data Preprocessing 2
What is the curse of dimensionality in machine learning?