WGU C951 TASK 3 | MACHINE LEARNING PROJECT
PROPOSAL FOR EMAIL SPAM FILTERING | 2026/27
UPDATE WITH COMPLETE SOLUTIONS.
130 Questions with Answers and Detailed Rationales
100 PERCENT GUARANTEED PASS
INSTANT DOWNLOAD ANSWERS INCLUDED
IMPORTANCE OF THIS DOCUMENT
This comprehensive examination preparation guide has been meticulously developed to help you succeed in the
WGU C951 TASK 3 | MACHINE LEARNING PROJECT PROPOSAL FOR EMAIL SPAM FILTERING | 2026/27
UPDATE WITH COMPLETE SOLUTIONS.. It contains 130 carefully selected questions that reflect the most
current exam content and testing strategies. Each question is accompanied by a correct answer and a detailed
rationale that explains the underlying pathophysiology, pharmacology, or clinical reasoning.
Self-Assessment – Test your knowledge and Exam Preparation – Familiarize yourself with the
identify areas requiring further question format and content
study areas
Concept Reinforcement – Deepen your Confidence Building – Develop test-taking
understanding through strategies and reduce
evidence-based exam anxiety
rationales
Time Management – Practice answering
questions under simulated
exam conditions
Review Summary 130 Questions
Foundations - Application - WGU C951 TASK 3 Machine Learning Project Proposal FOR Email SPAM
Filtering 2026/27 Update WITH Complete Solutions Machine Learning / DATA Science Graduate
All answers with rationales
,Table of Contents
Content Area Questions Key Topics
Machine Learning 1-22 Email, SPAM Filter, Project, Proposal, Model
Fundamentals
DATA Preprocessing AND 23-44 Emails, Filter, Proposal, False, Legitimate
Feature Engineering
Classification Algorithms 45-66 Model, Email, SPAM Filter, Proposal, Project
Model Evaluation AND 67-88 Model, Filter, Proposal, Email, False
Validation
Email SPAM Filtering 89-110 SPAM Filter, Proposal, Model, Email, Classifier
Techniques
Project Proposal 111-130 Email, Proposal, Filter, Model, Logistic Regression
Development
TOTAL 130 All questions include answers and detailed rationales
,Section A - Machine Learning Fundamentals
Q1.
A data scientist is building a spam filter for a multilingual user base. Emails are in English,
Spanish, and German. The team plans to use a pre-trained transformer model. Which
adaptation is most critical to avoid biased performance across languages?
A. Fine-tune the model on a combined B. Use a multilingual pre-trained model and
corpus of all languages without stratification evaluate performance separately per
language
C. Translate all emails to English before D. Train separate models for each language
classification and ensemble them
Correct: B - Use a multilingual pre-trained model and evaluate performance separately per
language
Rationale:Using a multilingual model and evaluating per language ensures that performance
is monitored and improved for each language, preventing bias. Fine-tuning without
stratification may favor the majority language. Translating to English loses nuance and may
not be feasible in real time. Separate models are complex and may not benefit from
cross-lingual transfer.
Q2.
In an email spam filter, the cost of a false negative (spam in inbox) is much higher than a
false positive (ham in spam folder). Which evaluation metric should the team optimize if
they want to directly incorporate this asymmetric cost into model selection?
A. Area under the ROC curve (AUC) B. F1 score
C. Expected cost, computed as a weighted D. Precision at a fixed recall of 0.95
sum of false positives and false negatives
Correct: C - Expected cost, computed as a weighted sum of false positives and false
negatives
Rationale:Expected cost directly incorporates the business costs of different error types,
allowing the model to be tuned to minimize total cost. AUC and F1 do not account for
asymmetric costs. Precision at a fixed recall only optimizes one operating point, not the full
trade-off.
Q3.
A team is using TF-IDF vectors for spam classification. They notice that some spam
emails contain very long gibberish text designed to dilute the spam keywords. Which
technique is most robust to this adversarial pattern?
Page 3
, Section A - Machine Learning Fundamentals
A. Normalize TF-IDF vectors to unit length B. Use sublinear TF scaling (e.g., 1 + log(tf))
C. Cap document length and truncate D. Use binary term presence instead of raw
emails to the first 500 words term frequency
Correct: B - Use sublinear TF scaling (e.g., 1 + log(tf))
Rationale:Sublinear scaling reduces the impact of very high term frequencies, so long
gibberish text doesn't dilute the signal as much. Normalization helps with document length but
not the dilution effect. Truncation may lose key spam signals. Binary presence ignores
frequency entirely, which may be too simplistic.
Q4.
A spam filter is trained on a dataset where 90% of emails are ham. The model achieves
98% accuracy but only 40% recall for spam. Which of the following is the most likely
cause and the best next step?
A. The model is overfitting; apply B. The dataset is imbalanced; use class
regularization weights or resampling
C. The features are not informative; use a D. The evaluation metric is wrong; use
neural network ROC-AUC instead
Correct: B - The dataset is imbalanced; use class weights or resampling
Rationale:The high accuracy with low spam recall is a classic symptom of class imbalance,
where the model predicts the majority class. Class weighting or resampling directly addresses
this. Regularization addresses overfitting, which is not indicated here. Feature engineering
may help, but the imbalance is the primary issue. ROC-AUC would also be high due to
imbalance, so it doesn't solve the problem.
Q5.
After deploying a spam filter, the team monitors its performance over time and notices a
gradual decline in precision. Which of the following is the most likely cause and the most
appropriate response?
A. Concept drift; retrain the model B. Data leakage; re-engineer the features
periodically with recent data
C. Overfitting; add dropout to the model D. The model is too slow; switch to a
simpler model
Correct: A - Concept drift; retrain the model periodically with recent data
Page 4
PROPOSAL FOR EMAIL SPAM FILTERING | 2026/27
UPDATE WITH COMPLETE SOLUTIONS.
130 Questions with Answers and Detailed Rationales
100 PERCENT GUARANTEED PASS
INSTANT DOWNLOAD ANSWERS INCLUDED
IMPORTANCE OF THIS DOCUMENT
This comprehensive examination preparation guide has been meticulously developed to help you succeed in the
WGU C951 TASK 3 | MACHINE LEARNING PROJECT PROPOSAL FOR EMAIL SPAM FILTERING | 2026/27
UPDATE WITH COMPLETE SOLUTIONS.. It contains 130 carefully selected questions that reflect the most
current exam content and testing strategies. Each question is accompanied by a correct answer and a detailed
rationale that explains the underlying pathophysiology, pharmacology, or clinical reasoning.
Self-Assessment – Test your knowledge and Exam Preparation – Familiarize yourself with the
identify areas requiring further question format and content
study areas
Concept Reinforcement – Deepen your Confidence Building – Develop test-taking
understanding through strategies and reduce
evidence-based exam anxiety
rationales
Time Management – Practice answering
questions under simulated
exam conditions
Review Summary 130 Questions
Foundations - Application - WGU C951 TASK 3 Machine Learning Project Proposal FOR Email SPAM
Filtering 2026/27 Update WITH Complete Solutions Machine Learning / DATA Science Graduate
All answers with rationales
,Table of Contents
Content Area Questions Key Topics
Machine Learning 1-22 Email, SPAM Filter, Project, Proposal, Model
Fundamentals
DATA Preprocessing AND 23-44 Emails, Filter, Proposal, False, Legitimate
Feature Engineering
Classification Algorithms 45-66 Model, Email, SPAM Filter, Proposal, Project
Model Evaluation AND 67-88 Model, Filter, Proposal, Email, False
Validation
Email SPAM Filtering 89-110 SPAM Filter, Proposal, Model, Email, Classifier
Techniques
Project Proposal 111-130 Email, Proposal, Filter, Model, Logistic Regression
Development
TOTAL 130 All questions include answers and detailed rationales
,Section A - Machine Learning Fundamentals
Q1.
A data scientist is building a spam filter for a multilingual user base. Emails are in English,
Spanish, and German. The team plans to use a pre-trained transformer model. Which
adaptation is most critical to avoid biased performance across languages?
A. Fine-tune the model on a combined B. Use a multilingual pre-trained model and
corpus of all languages without stratification evaluate performance separately per
language
C. Translate all emails to English before D. Train separate models for each language
classification and ensemble them
Correct: B - Use a multilingual pre-trained model and evaluate performance separately per
language
Rationale:Using a multilingual model and evaluating per language ensures that performance
is monitored and improved for each language, preventing bias. Fine-tuning without
stratification may favor the majority language. Translating to English loses nuance and may
not be feasible in real time. Separate models are complex and may not benefit from
cross-lingual transfer.
Q2.
In an email spam filter, the cost of a false negative (spam in inbox) is much higher than a
false positive (ham in spam folder). Which evaluation metric should the team optimize if
they want to directly incorporate this asymmetric cost into model selection?
A. Area under the ROC curve (AUC) B. F1 score
C. Expected cost, computed as a weighted D. Precision at a fixed recall of 0.95
sum of false positives and false negatives
Correct: C - Expected cost, computed as a weighted sum of false positives and false
negatives
Rationale:Expected cost directly incorporates the business costs of different error types,
allowing the model to be tuned to minimize total cost. AUC and F1 do not account for
asymmetric costs. Precision at a fixed recall only optimizes one operating point, not the full
trade-off.
Q3.
A team is using TF-IDF vectors for spam classification. They notice that some spam
emails contain very long gibberish text designed to dilute the spam keywords. Which
technique is most robust to this adversarial pattern?
Page 3
, Section A - Machine Learning Fundamentals
A. Normalize TF-IDF vectors to unit length B. Use sublinear TF scaling (e.g., 1 + log(tf))
C. Cap document length and truncate D. Use binary term presence instead of raw
emails to the first 500 words term frequency
Correct: B - Use sublinear TF scaling (e.g., 1 + log(tf))
Rationale:Sublinear scaling reduces the impact of very high term frequencies, so long
gibberish text doesn't dilute the signal as much. Normalization helps with document length but
not the dilution effect. Truncation may lose key spam signals. Binary presence ignores
frequency entirely, which may be too simplistic.
Q4.
A spam filter is trained on a dataset where 90% of emails are ham. The model achieves
98% accuracy but only 40% recall for spam. Which of the following is the most likely
cause and the best next step?
A. The model is overfitting; apply B. The dataset is imbalanced; use class
regularization weights or resampling
C. The features are not informative; use a D. The evaluation metric is wrong; use
neural network ROC-AUC instead
Correct: B - The dataset is imbalanced; use class weights or resampling
Rationale:The high accuracy with low spam recall is a classic symptom of class imbalance,
where the model predicts the majority class. Class weighting or resampling directly addresses
this. Regularization addresses overfitting, which is not indicated here. Feature engineering
may help, but the imbalance is the primary issue. ROC-AUC would also be high due to
imbalance, so it doesn't solve the problem.
Q5.
After deploying a spam filter, the team monitors its performance over time and notices a
gradual decline in precision. Which of the following is the most likely cause and the most
appropriate response?
A. Concept drift; retrain the model B. Data leakage; re-engineer the features
periodically with recent data
C. Overfitting; add dropout to the model D. The model is too slow; switch to a
simpler model
Correct: A - Concept drift; retrain the model periodically with recent data
Page 4
