Lasso regularization can lead to sparse solutions where some coefficients become exactly ________.
- Negative
- Positive
- Zero
- nan
Lasso regularization adds an L1 penalty, causing some coefficients to be exactly zero, leading to sparsity in the model's solution.
What are the potential risks of using too high a degree in Polynomial Regression?
- Decreased complexity
- Increased bias
- Increased variance and overfitting
- Simplified model
Using too high a degree in Polynomial Regression can lead to increased variance and overfitting. It makes the model too complex, fitting the noise in the training data, and thus failing to generalize well to unseen data.
How is overfitting specifically related to Polynomial Regression?
- It's not related
- Overfitting can occur with high-degree polynomials
- Overfitting only happens with linear models
- Polynomial Regression prevents overfitting
Overfitting in Polynomial Regression can occur when using high-degree polynomials. The model becomes too flexible and fits the training data too well, including the noise, which reduces its ability to generalize well to unseen data.
What is the significance of the slope in Simple Linear Regression?
- It Describes the Rate of Change in Y for a One-Unit Change in X
- It Indicates the Intercept
- It Predicts the Error
- It Shows the Starting Point of the Line
The slope in Simple Linear Regression describes the rate of change in the dependent variable (Y) for a one-unit change in the independent variable (X).
You have developed a regression model, and the R-Squared value is very close to 1. What could this indicate, and what would you check?
- Good fit; No need to check anything
- Perfect fit; Check for overfitting
- Perfect fit; Check for underfitting
- Poor fit; Check for bias
An R-Squared value close to 1 typically indicates a nearly perfect fit, but this might be a sign of overfitting. It is essential to verify the model's performance on unseen data, as it may be capturing noise and specificities of the training data rather than the underlying trend. Cross-validation or a hold-out validation set can help in this assessment.
You are building a model to predict whether a given email is spam or not. Why might Logistic Regression be a suitable approach?
- Because it can model binary outcomes and estimate probabilities
- Because it can predict multiple classes
- Because it works well with unstructured data
- Because it's a regression algorithm
Logistic Regression is suitable for binary classification problems such as spam detection, as it models binary outcomes and can estimate the probability of an email being spam or not.
You've built a model with high variance. How can Cross-Validation help in diagnosing and improving the model?
- By automatically reducing the complexity of the model
- By helping in feature selection
- By providing a robust estimation of model performance and aiding hyperparameter tuning
- By providing more data for training
Cross-Validation provides a robust estimation of the model's performance across different data splits. For a high variance model, it can help in diagnosing the issue by highlighting overfitting and assist in hyperparameter tuning to find the best complexity that captures underlying patterns without fitting noise.
Describe how the concepts of features, targets, training, and testing are interrelated in Machine Learning.
- Features and targets are for clustering; Training and testing for prediction
- Features and targets are unrelated; Training and testing are used interchangeably
- Features are for prediction; Targets for evaluation; Training and testing are unrelated
- Features are used to predict targets; Training is learning patterns; Testing evaluates performance
Features are the input variables used to predict targets. Training involves learning the patterns from features to predict targets, and testing evaluates how well this learning generalizes to unseen data. These concepts are essential in building and evaluating supervised learning models.
You have a dataset with hundreds of features, some of which are redundant. How would you approach reducing the dimensionality?
- Remove all redundant features manually
- Apply PCA
- Use only the first few features
- Normalize the data
Applying Principal Component Analysis (PCA) would be the most efficient way to reduce dimensionality in this scenario. PCA transforms the data into a new set of uncorrelated features, effectively capturing the most important variance in fewer dimensions, and thus removing redundancy. Manually removing redundant features may not be practical with hundreds of features, and other options do not directly address dimensionality reduction.
Which learning paradigm does not require labeled data and finds hidden patterns in the data?
- Reinforcement Learning
- Semi-supervised Learning
- Supervised Learning
- Unsupervised Learning
Unsupervised Learning does not require labeled data and works by finding hidden patterns and structures in the data.