Week2_S1

ROC curves plot which relationship?

• Precision vs. Recall

• Recall vs. Specificity

• True Positive Rate vs. False Positive Rate

• Precision vs. Accuracy

SVMs aim to maximize the __________ between classes.

Random forest prediction for classification is based on:

• Single best tree

• Weighted averaging of features

• Averaging probabilities across trees

•  Taking the majority vote

A decision tree split that produces pure child nodes has:

• High entropy

•  Low entropy

• No information gain

• Random splits

Pruning a Decision Tree helps to:

•  Increase depth

• Reduce overfitting

• Increase overfitting

• Increase entropy

Which of the following explains why accuracy can be misleading for imbalanced datasets

• Accuracy equally weighs all errors.
  

• Accuracy is always better than precision/recall.

• Classifiers can achieve high accuracy by predicting the majority class only.

• Accuracy automatically adjusts for skewed classes

In logistic regression, the cost function commonly used is

• Sigmoid

•  Cross-entropy loss

• Mean squared error

• Hinge loss

The Perceptron differs from Logistic regression by using the ______ activation instead of the sigmoid.

Which of the following statements correctly describes the relationship between biological and artificial neural networks?

• Artificial neural networks directly replicate the exact biological processes of the human brain.

• Biological neurons can be trained using gradient descent like artificial neurons.

• Artificial neurons are connected through synapses, just like biological neurons.

• Artificial neural networks simulate the learning mechanism of biological neurons using computational units and weights.

why has deep learning recently surpassed traditional machine learning methods in many tasks?

• Because deep learning completely eliminates the need for feature engineering.

• Because deep learning algorithms can solve more complex functions.

• Due to advancements in data availability, computational power, and experimentation.

• Because neural networks require less training data than traditional methods.