Recap Quiz

In the CRISP-DM methodology, what is the primary purpose of the Data Understanding phase?

• To clean and transform raw data into a usable format

• To explore and gather initial insights about the data through tables and visualizations

• To build and evaluate machine learning models

• To deploy the final model into production

During data cleaning, why might you choose to clamp outliers rather than remove those rows entirely?

• Clamping is required by most machine learning algorithms.

• Clamping is faster computationally than deleting rows.

• Clamping preserves all data points and reduces the influence of extremes without losing information about those cases.

• Removing rows corrupts the original dataset permanently.

When preparing data for a machine learning model, why is stratified sampling important when splitting your data into training and test sets?

• It ensures both the training and test sets have similar class distributions for the target variable.

• It guarantees equal numbers of every class in each set.

• It automatically normalizes the numeric features.

• It prevents the model from overfitting to the majority class.

A basketball player's weight is standardized using a z-score transformation. If a player's weight is exactly equal to the mean weight of the dataset, what will their standardized value be?

• -1

• 0

• 1

• 0.5

A linear regression model evaluates its predictions and achieves a Mean Absolute Error (MAE) of 25 and a Root Mean Squared Error (RMSE) of 45. Why is the RMSE noticeably larger than the MAE?

• MAE cannot exceed RMSE by definition.

• RMSE uses entirely different units of measurement than MAE.

• There was likely a calculation error during evaluation.

• RMSE penalizes large errors more heavily due to the squaring of the residuals.

What is the primary output of a logistic regression model before any decision threshold is applied?

• A probability score for class membership

• A definitive class label (e.g., Class 0 or Class 1)

• The distance to the nearest cluster centroid

• A continuous numeric prediction with no upper bound

In an imbalanced classification setting (e.g., a dataset with 95% negative cases and 5% positive cases), why can evaluating the model solely on "Accuracy" be misleading?

• Accuracy cannot be computed for binary classification tasks.

• Accuracy ignores class distribution and a model can appear highly accurate simply by predicting the majority class every time.

• Accuracy requires probabilities rather than discrete labels.

• Accuracy is only a valid metric when using cross-validation.

In a medical screening context where failing to detect a disease is highly dangerous, which type of error is usually the most critical to minimize?

• False Positive

• False Negative

• True Positive

• True Negative

Which parameter is commonly used to help prevent a single Decision Tree from overfitting the training data?

• Increasing the number of features

• Removing feature scaling

• Setting a maximum tree depth

• Using more categorical variables

A Random Forest builds multiple trees, each trained on a bootstrap sample. In this context, what is a bootstrap sample?

• A stratified subset that guarantees an exact 50/50 class balance.

• A random sample drawn with replacement from the training data.

• A random sample drawn without replacement from the training data.

• A subset of features rather than a subset of training examples.