How does Logistic Regression differ from Linear Regression?
- Both are used for clustering
- Models binary outcomes
- No difference
- Uses the same equations
While Linear Regression is used to predict continuous outcomes, Logistic Regression is used to model binary outcomes. They use different mathematical functions to accomplish this.
You have two datasets: one with 10,000 instances and one with 100,000. How would the size of the datasets impact the choice of Machine Learning algorithm?
- Larger datasets always require more complex models
- Larger datasets may require more careful consideration of computational efficiency
- Size doesn’t impact the choice of algorithm
- Size only impacts the choice of hyperparameters
Larger datasets may require more computational resources, so choosing an algorithm that can handle the computational complexity might be critical. The choice might not be solely determined by the size, but it can certainly impact the decision.
What is the statistical interpretation of interaction effects in Multiple Linear Regression?
- Combined effect of two variables affecting the dependent variable
- Minimizing overfitting
- Removing irrelevant features
- Transforming variables into linear space
Interaction effects statistically represent the combined effect of two or more variables on the dependent variable. They capture relationships that are not apparent when variables are considered separately.
You are faced with a multi-class classification problem. How would the choice of K and distance metric affect the KNN algorithm's ability to differentiate between the classes?
- Choice of K affects precision, distance metric affects generalization
- Choice of K affects recall, distance metric affects speed
- Choice of K and distance metric carefully affects differentiation between classes
- It has no effect
The careful selection of K and distance metric can greatly affect the KNN algorithm's ability to differentiate between classes in multi-class classification.
In the context of K-Means clustering, what challenges may arise with poorly initialized centroids?
- Faster convergence
- No convergence
- No effect on clustering
- Suboptimal clustering, Slow convergence
Poorly initialized centroids in K-Means may lead to suboptimal clustering and slow convergence. If the centroids are initialized very poorly, it might even cause the algorithm to get stuck in local minima.
When a Polynomial Regression model is overfitted, it will perform well on the _________ data but poorly on the _________ data.
- test, training
- training, test
- training, validation
- validation, training
An overfitted Polynomial Regression model will fit the training data very well but perform poorly on unseen test data.
You are asked to include an interaction effect between two variables in a Multiple Linear Regression model. How would you approach this task, and what considerations would you need to keep in mind?
- Add the variables
- Divide the variables
- Multiply the variables and include the interaction term in the model
- Multiply the variables together
Including an interaction effect involves multiplying the variables together and adding this interaction term to the model. It's important to consider the meaningfulness of the interaction, possible multicollinearity with other variables, and the potential need for centering the variables to minimize issues with interpretation.
In what scenarios would you use PCA, and when would you opt for other methods like LDA or t-SNE?
- Use PCA for high-dimensional data, LDA for linearly separable, t-SNE for non-linear
- Use PCA for labeled data, LDA for unlabeled, t-SNE for large-scale
- Use PCA for large-scale, LDA for visualization, t-SNE for labeled data
- Use PCA for noisy data, LDA for small-scale, t-SNE for visualizations
Use PCA when dealing with high-dimensional data and the primary goal is to reduce dimensions by maximizing variance. LDA is suitable when class labels are available, and the data is linearly separable. t-SNE is often used for non-linear data and is especially useful for visualizations, as it preserves local structures.
In a high-dimensional dataset, how would you decide which kernel to use for SVM?
- Always use RBF kernel
- Always use linear kernel
- Choose the kernel randomly
- Use cross-validation to select the best kernel
By using cross-validation, you can compare different kernels' performance and choose the one that gives the best validation accuracy.
You're designing a self-driving car's navigation system. How would reinforcement learning be applied in this context?
- To cluster traffic patterns
- To combine labeled and unlabeled data
- To learn optimal paths through rewards/penalties
- To use only labeled data for navigation
Reinforcement Learning would enable the navigation system to learn optimal paths by interacting with the environment and receiving feedback through rewards and penalties.
What are some common techniques to avoid overfitting?
- Increasing model complexity, Adding noise, Cross-validation
- Increasing model complexity, Regularization, Cross-validation
- Reducing model complexity, Adding noise, Cross-validation
- Reducing model complexity, Regularization, Cross-validation
Common techniques to avoid overfitting include "reducing model complexity, regularization, and cross-validation." These methods prevent the model from fitting too closely to the training data.
After applying PCA to your dataset, you find that some Eigenvectors have very small corresponding Eigenvalues. What does this indicate, and what action might you take?
- This indicates a problem with the data and you must discard it
- This indicates that these eigenvectors capture little variance, and you may choose to discard them
- This is an indication that PCA is not suitable for your data
- This means that you must include these eigenvectors
Very small eigenvalues indicate that the corresponding eigenvectors capture little variance, and discarding them would reduce dimensions without losing much meaningful information.