Machine Learning Practice Questions: Test Your Knowledge | LearnByTeaching.ai
Assess your machine learning understanding with these 40 questions covering supervised learning, unsupervised learning, neural networks, and model evaluation. These questions test both theoretical foundations and practical intuition needed to build effective ML systems.
40 questions total
Supervised Learning
Test your understanding of supervised learning.
In linear regression, the cost function typically minimized is:
The bias-variance tradeoff implies that a model with very high complexity will likely have:
L2 regularization (Ridge regression) adds what term to the loss function?
A decision tree splits data at each node by:
The kernel trick in SVMs allows the algorithm to:
Logistic regression outputs:
In k-nearest neighbors (KNN), increasing k generally:
Naive Bayes is called 'naive' because it assumes:
Random Forest improves over a single decision tree by:
Gradient boosting builds an ensemble by:
Unsupervised Learning
Test your understanding of unsupervised learning.
K-means clustering requires the user to specify in advance:
Principal Component Analysis (PCA) finds directions that:
DBSCAN differs from K-means in that DBSCAN:
In hierarchical clustering, a dendrogram shows:
The elbow method for choosing k in K-means plots:
t-SNE is primarily used for:
Gaussian Mixture Models (GMM) extend K-means by:
Autoencoders learn representations by:
The silhouette score measures:
Association rule mining (e.g., Apriori algorithm) is commonly used in:
Neural Networks and Deep Learning
Test your understanding of neural networks and deep learning.
The vanishing gradient problem in deep networks causes:
The ReLU activation function is defined as:
Dropout regularization works by:
Convolutional Neural Networks (CNNs) are particularly effective for image data because:
In backpropagation, the gradient of the loss with respect to each weight is computed using:
Batch normalization helps training by:
A recurrent neural network (RNN) is designed for:
The Transformer architecture relies primarily on:
Transfer learning involves:
GANs (Generative Adversarial Networks) consist of:
Model Evaluation and Practical ML
Test your understanding of model evaluation and practical ml.
K-fold cross-validation works by:
Precision in binary classification measures:
The ROC curve plots:
Data leakage occurs when:
Feature scaling (standardization or normalization) is important for which algorithms?
The F1 score is the:
When dealing with a highly imbalanced dataset (99% negative, 1% positive), accuracy is a poor metric because:
Hyperparameter tuning using grid search:
One-hot encoding is used to convert:
Early stopping is a regularization technique that:
Scoring Guide
Total possible: 40
Study Recommendations
- Implement linear regression, logistic regression, and a simple neural network from scratch in NumPy
- Practice explaining the bias-variance tradeoff and when to use which regularization technique
- Build end-to-end ML projects with real messy data to develop practical intuition
- Study the math behind backpropagation with pen and paper before relying on frameworks
- Use the teach-back method — explain each algorithm's assumptions, strengths, and limitations as if teaching a colleague
More Machine Learning Resources
Want more machine learning practice?
Upload your notes and LearnByTeaching.ai generates unlimited practice questions tailored to your course. Then teach the concepts to AI students who challenge your understanding.
Try LearnByTeaching.ai — It's Free