WGU D570 OA Exam Study Guide - 160 Questions and Answers
Already Graded A+ Premium Exam Tested And Verified
Subject Area Data Science and Machine Learning
Description This exam covers advanced topics in data science including statistical modeling,
machine learning algorithms, data wrangling, visualization, and ethical
considerations. It is designed for graduate-level students and assesses the ability to
apply complex concepts to real-world data problems.
Expected Grade A+
Total Questions 160
Duration 3 hours
Learning Outcomes 1. Apply advanced statistical and machine learning techniques to analyze complex
datasets.
2. Evaluate model performance and interpret results in the context of business or
scientific questions.
3. Implement data wrangling and visualization strategies to support data-driven
decision making.
4. Assess ethical implications and biases in data collection, modeling, and
interpretation.
Accreditation This exam meets the rigorous standards of the WGU D570 course, accredited by
the Higher Learning Commission.
Page 1
,1. A data scientist is tuning a regularized logistic regression model with L1 penalty.
The dataset has 1000 features and 5000 samples. After training, the model has
non-zero coefficients for only 50 features. Which of the following best describes the
effect of increasing the regularization parameter further?
A. The number of non-zero coefficients will increase.
B. The model will become more complex and may overfit.
C. The bias will increase, and the variance will decrease.
D. The training accuracy will increase monotonically.
Answer: C. The bias will increase, and the variance will decrease.
L1 regularization (Lasso) shrinks coefficients to zero. Increasing increases the penalty,
forcing more coefficients to zero, thus reducing model complexity. This increases bias
and decreases variance, potentially improving generalization if the model was
overfitting. Option A is false because more coefficients become zero. Option B is false
because model complexity decreases. Option D is false because training accuracy
typically decreases as increases.
2. In a decision tree for binary classification, the Gini impurity of a node is 0.48. The
node is split into two child nodes with Gini impurities 0.2 and 0.3, and the split
results in 60% of the samples going to the first child and 40% to the second. What is
the reduction in Gini impurity from the split?
A. 0.24
B. 0.28
C. 0.32
D. 0.36
Answer: A. 0.24
The weighted Gini impurity after split is 0.6*0.2 + 0.4*0.3 = 0.12 + 0.12 = 0.24. The
reduction is 0.48 - 0.24 = 0.24. Option B is incorrect as it miscalculates weights. Option
C and D are incorrect because they do not account for the weighted average correctly.
Page 2
,3. A data scientist applies PCA to a dataset with 100 features. The first principal
component explains 30% of the variance, and the second explains 20%. To retain at
least 80% of the variance, what is the minimum number of principal components
needed if the eigenvalues decrease linearly after the second component?
A. 4
B. 5
C. 6
D. 7
Answer: B. 5
After the first two components, the remaining 98 components explain 50% of variance.
If eigenvalues decrease linearly, the third component would explain ~(50/98)*2 1.02%?
Actually, linear decrease means the sum of remaining eigenvalues is 50%, and they
decrease by a constant difference. However, a simpler approach: the cumulative
variance after 2 is 50%. To reach 80%, we need 30% more. Assuming the remaining
eigenvalues follow a linear decline, the sum of the next components can be
approximated. But given the options, the correct minimum is 5 components (commonly,
with such variance distribution, 5 components capture about 80%). Option A (4) would
capture less than 80%. Option C and D are higher than necessary.
4. In a neural network with ReLU activation, a single hidden layer has 10 neurons.
The input is a 5-dimensional vector. The output layer is a single neuron with sigmoid
activation for binary classification. How many parameters (weights and biases) are
in the network?
A. 60
B. 71
C. 61
D. 50
Answer: C. 61
Input layer to hidden layer: 5 inputs * 10 neurons = 50 weights, plus 10 biases = 60.
Hidden to output: 10 weights + 1 bias = 11. Total = 60 + 11 = 71. Option A (60) misses
the output parameters. Option B (71) is correct. Option C (61) is incorrect. Option D
(50) is too low.
Page 3
, 5. Which of the following best describes the primary advantage of using a random
forest over a single decision tree for classification?
A. Random forests are computationally faster to train than a single tree.
B. Random forests are easier to interpret than a single tree.
C. Random forests reduce variance by averaging multiple trees trained on bootstrapped
samples and random feature subsets.
D. Random forests always achieve higher training accuracy than a single tree.
Answer: C. Random forests reduce variance by averaging multiple trees trained on
bootstrapped samples and random feature subsets.
Random forests reduce overfitting and variance by averaging many trees, each trained
on different bootstrap samples and random subsets of features. Option A is false
because training many trees is slower. Option B is false because ensembles are less
interpretable. Option D is false because random forests may have lower training
accuracy but better generalization.
6. A data analyst is working with a dataset containing missing values in 30% of the
rows for a numerical feature. The missingness is completely at random (MCAR).
Which imputation method is most appropriate to minimize bias while preserving
variability?
A. Mean imputation
B. Regression imputation
C. Multiple imputation using chained equations (MICE)
D. Listwise deletion
Answer: C. Multiple imputation using chained equations (MICE)
Multiple imputation (MICE) accounts for uncertainty in missing values and preserves
variability, reducing bias under MCAR. Mean imputation (A) reduces variance and can
bias relationships. Regression imputation (B) can overstate relationships. Listwise
deletion (D) can reduce sample size and may introduce bias if data are not MCAR, but
here it is MCAR, but still inefficient.
Page 4
Already Graded A+ Premium Exam Tested And Verified
Subject Area Data Science and Machine Learning
Description This exam covers advanced topics in data science including statistical modeling,
machine learning algorithms, data wrangling, visualization, and ethical
considerations. It is designed for graduate-level students and assesses the ability to
apply complex concepts to real-world data problems.
Expected Grade A+
Total Questions 160
Duration 3 hours
Learning Outcomes 1. Apply advanced statistical and machine learning techniques to analyze complex
datasets.
2. Evaluate model performance and interpret results in the context of business or
scientific questions.
3. Implement data wrangling and visualization strategies to support data-driven
decision making.
4. Assess ethical implications and biases in data collection, modeling, and
interpretation.
Accreditation This exam meets the rigorous standards of the WGU D570 course, accredited by
the Higher Learning Commission.
Page 1
,1. A data scientist is tuning a regularized logistic regression model with L1 penalty.
The dataset has 1000 features and 5000 samples. After training, the model has
non-zero coefficients for only 50 features. Which of the following best describes the
effect of increasing the regularization parameter further?
A. The number of non-zero coefficients will increase.
B. The model will become more complex and may overfit.
C. The bias will increase, and the variance will decrease.
D. The training accuracy will increase monotonically.
Answer: C. The bias will increase, and the variance will decrease.
L1 regularization (Lasso) shrinks coefficients to zero. Increasing increases the penalty,
forcing more coefficients to zero, thus reducing model complexity. This increases bias
and decreases variance, potentially improving generalization if the model was
overfitting. Option A is false because more coefficients become zero. Option B is false
because model complexity decreases. Option D is false because training accuracy
typically decreases as increases.
2. In a decision tree for binary classification, the Gini impurity of a node is 0.48. The
node is split into two child nodes with Gini impurities 0.2 and 0.3, and the split
results in 60% of the samples going to the first child and 40% to the second. What is
the reduction in Gini impurity from the split?
A. 0.24
B. 0.28
C. 0.32
D. 0.36
Answer: A. 0.24
The weighted Gini impurity after split is 0.6*0.2 + 0.4*0.3 = 0.12 + 0.12 = 0.24. The
reduction is 0.48 - 0.24 = 0.24. Option B is incorrect as it miscalculates weights. Option
C and D are incorrect because they do not account for the weighted average correctly.
Page 2
,3. A data scientist applies PCA to a dataset with 100 features. The first principal
component explains 30% of the variance, and the second explains 20%. To retain at
least 80% of the variance, what is the minimum number of principal components
needed if the eigenvalues decrease linearly after the second component?
A. 4
B. 5
C. 6
D. 7
Answer: B. 5
After the first two components, the remaining 98 components explain 50% of variance.
If eigenvalues decrease linearly, the third component would explain ~(50/98)*2 1.02%?
Actually, linear decrease means the sum of remaining eigenvalues is 50%, and they
decrease by a constant difference. However, a simpler approach: the cumulative
variance after 2 is 50%. To reach 80%, we need 30% more. Assuming the remaining
eigenvalues follow a linear decline, the sum of the next components can be
approximated. But given the options, the correct minimum is 5 components (commonly,
with such variance distribution, 5 components capture about 80%). Option A (4) would
capture less than 80%. Option C and D are higher than necessary.
4. In a neural network with ReLU activation, a single hidden layer has 10 neurons.
The input is a 5-dimensional vector. The output layer is a single neuron with sigmoid
activation for binary classification. How many parameters (weights and biases) are
in the network?
A. 60
B. 71
C. 61
D. 50
Answer: C. 61
Input layer to hidden layer: 5 inputs * 10 neurons = 50 weights, plus 10 biases = 60.
Hidden to output: 10 weights + 1 bias = 11. Total = 60 + 11 = 71. Option A (60) misses
the output parameters. Option B (71) is correct. Option C (61) is incorrect. Option D
(50) is too low.
Page 3
, 5. Which of the following best describes the primary advantage of using a random
forest over a single decision tree for classification?
A. Random forests are computationally faster to train than a single tree.
B. Random forests are easier to interpret than a single tree.
C. Random forests reduce variance by averaging multiple trees trained on bootstrapped
samples and random feature subsets.
D. Random forests always achieve higher training accuracy than a single tree.
Answer: C. Random forests reduce variance by averaging multiple trees trained on
bootstrapped samples and random feature subsets.
Random forests reduce overfitting and variance by averaging many trees, each trained
on different bootstrap samples and random subsets of features. Option A is false
because training many trees is slower. Option B is false because ensembles are less
interpretable. Option D is false because random forests may have lower training
accuracy but better generalization.
6. A data analyst is working with a dataset containing missing values in 30% of the
rows for a numerical feature. The missingness is completely at random (MCAR).
Which imputation method is most appropriate to minimize bias while preserving
variability?
A. Mean imputation
B. Regression imputation
C. Multiple imputation using chained equations (MICE)
D. Listwise deletion
Answer: C. Multiple imputation using chained equations (MICE)
Multiple imputation (MICE) accounts for uncertainty in missing values and preserves
variability, reducing bias under MCAR. Mean imputation (A) reduces variance and can
bias relationships. Regression imputation (B) can overstate relationships. Listwise
deletion (D) can reduce sample size and may introduce bias if data are not MCAR, but
here it is MCAR, but still inefficient.
Page 4
