WorksheetsMachine Learning Quiz
Total questions: 30
Worksheet time: 15mins
What is the main goal of Supervised Learning?
To find hidden patterns in data without labels.
To make predictions based on labeled input and output data.
To improve a model through trial and error in an environment.
To group similar data points together.
In the equation for a linear model, f(x)=W⋅X+b, what does 'W' represent?
The input data.
The prediction.
The bias.
The weight.
The Cost Function in machine learning is used to measure what?
The time it takes to train the model.
The difference between the model's predictions and the actual values.
The number of features in the dataset.
The speed of the computer's processor.
Which algorithm is primarily used for regression tasks where the output is a continuous value?
K-Means Clustering
Logistic Regression
Linear Regression
PCA
What is the primary purpose of the Gradient Descent algorithm?
To visualize the data.
To clean the data by removing missing values.
To update the model's weights and biases to minimize the cost function.
To split the data into training and testing sets.
A Decision Tree makes predictions by:
Fitting a single straight line to the data.
Creating a complex mathematical equation.
Following a tree-like structure of step-by-step decisions.
Averaging the predictions of many different models.
For which of the following scenarios would a Decision Tree be more suitable than Linear Regression?
Predicting house prices based on square footage.
When the relationship between variables is perfectly linear.
When the data has complex, non-linear patterns and interactions.
When you have only one feature and one label.
In a Decision Tree used for classifying if a fruit is an 'Apple' or 'Orange', what would a 'node' represent?
The final prediction ('Apple').
A feature used to make a decision, like 'Color' or 'Size'.
The accuracy of the model.
The entire dataset of fruits.
Can a Decision Tree be used for both classification and regression problems?
No, it is only for classification.
No, it is only for regression.
Yes, it can be used for both.
Only if the data has no missing values.
Is it possible to create many different-looking Decision Trees from the exact same dataset?
No, only one unique tree can ever be made.
Yes, many different trees can be constructed depending on how the features are split.
Only if you add more data.
Only if you use a different algorithm.
What is the core idea behind 'Ensemble Learning'?
Using only one, highly complex model.
Focusing exclusively on data cleaning.
Combining the predictions from several weaker models to create one strong model.
A method for visualizing high-dimensional data.
Random Forest is a popular ensemble method. What is it an ensemble of?
Linear Regression models.
K-Means clusters.
Logistic Regression models.
Decision Trees.
In 'Boosting,' how does the algorithm train the sequence of models?
All models are trained simultaneously and independently.
Each new model is trained to correct the errors made by the previous model.
It randomly selects one model to be the final predictor.
It trains models on different, random subsets of the features.
The 'Bagging' technique primarily helps to:
Increase the model's bias.
Reduce the model's variance and prevent overfitting.
Speed up the data loading process.
Create new features from existing ones.
Besides accuracy, what is another important factor to consider when choosing between a Random Forest and a Gradient Boosting model for a real-world application?
The color of the graphs they produce.
The programming language used.
Training time, prediction speed, or model size.
The name of the dataset.
In machine learning, what does 'generalization' refer to?
The model's ability to perform well only on the data it was trained on.
The model's ability to perform well on new, unseen data.
The time it takes to train a model.
The process of creating new features.
The 'Bias-Variance Tradeoff' describes the challenge of:
Choosing between Python and R.
Finding a model that is not too simple (underfitting) but not too complex (overfitting).
Splitting data into training and testing sets.
Deciding whether to use supervised or unsupervised learning.
A model with very high bias is likely to:
Fit the training data perfectly and fail on new data.
Be too simple and fail to capture the underlying patterns in the data (underfitting).
Have too many features.
Be impossible to train.
According to Pedro Domingos's influential paper, what is often the key to creating a successful model?
Trying every single algorithm possible.
Spending most of your time on clever feature engineering.
Using the most complicated model you can find.
Getting more training data, even if it's low quality.
Creating a FamilySize feature from the SibSp and Parch columns in the Titanic dataset is a perfect example of:
Model Optimization.
Data Evaluation.
Feature Engineering.
Model Representation.
What is the main difference between Supervised and Unsupervised Learning?
Supervised Learning uses numerical data, while Unsupervised uses text.
Unsupervised Learning works with data that has no 'y' labels or answer key.
Supervised Learning is always more accurate.
Unsupervised Learning is only used for reinforcement learning.
The K-Means algorithm is a popular method for what type of task?
Regression.
Classification.
Clustering.
Dimensionality Reduction.
In K-Means, how do you find the best value for 'K' (the number of clusters)?
Always choose K=3.
Use the Elbow Method to find where the error rate stops decreasing sharply.
Pick the highest number of clusters possible.
The value of K is always the number of features.
Why is it important to scale your features (e.g., using StandardScaler) before running K-Means?
It is not important and has no effect on the outcome.
Scaling is only necessary for Decision Trees.
To prevent features with large values from unfairly dominating the distance calculations.
To convert all numbers into text.
In the K-Means algorithm, after assigning all data points to a cluster, what is the next step?
The algorithm stops.
All data points are deleted.
The centroid of each cluster is moved to the average position of the points in its cluster.
A new feature is created.
What is the primary goal of Principal Component Analysis (PCA)?
To make a final prediction for a classification problem.
To reduce the number of features (dimensionality) in a dataset while keeping the most important information.
To find the best learning rate for gradient descent.
To group similar customers together.
Using PCA to reduce 784 features (from an image) down to just 2 features is useful for:
Making the model more complex.
Making the final prediction more accurate.
Increasing the time it takes to train the model.
Visualizing the data in a 2D scatter plot and speeding up other algorithms.
In a business context, like segmenting mall customers, what is the most important final step after running a clustering algorithm?
Reporting the final value of the cost function.
Deleting the original data.
Translating the technical clusters into meaningful business 'personas' or insights.
Re-running the model with more clusters.
If a K-Means model identifies a customer segment with high income but low spending, what kind of 'persona' would this be?
A target for premium, high-value offers.
A 'Careful Spender' or 'Untapped Potential' that marketing could investigate.
The average, typical customer.
A frugal customer who is not worth marketing to.
K-Means has a limitation where it assumes clusters are spherical. What kind of cluster shape would it struggle with?
A perfect circle.
A dense, round group of points.
A long, thin, or irregularly shaped cluster.
A cluster with very few data points.
