WGU D600 - Statistical Data Mining | Task 2
WGU D600
Statistical Data Mining
TASK 2: MULTIPLE LOGISTIC REGRESSION ANALYSIS
Completed Performance Assessment Model | September 2026
Predicting Telecommunications Customer Churn with Multiple Logistic
Regression
Scope and current alignment
WGU identifies D600 as Statistical Data Mining in the current September 2026 catalog. The course focuses on
regression plus supervised classification and prediction. Publicly documented D600 workflow identifies Task
2 as logistic regression and describes version-controlled code, train/test evaluation, accuracy and confusion-
matrix evidence, model reduction, and a presentation. The complete live learner-portal rubric and assigned
dataset are not publicly available, so this document uses a representative telecommunications churn
scenario and explicitly avoids presenting invented material as proprietary WGU content.
Original academic model. It is not an official WGU exemplar and no evaluator-issued pass is claimed.
September 2026 aligned performance assessment model | Page 1
, WGU D600 - Statistical Data Mining | Task 2
Executive Summary
This analysis addresses a practical retention question: which customer attributes are associated with
churn, and how accurately can a multiple logistic regression model classify whether a customer will
leave? A representative dataset of 1,800 customer records was analyzed. The binary outcome Churn
equals 1 for a customer who leaves and 0 for a customer who remains. The initial model included five
continuous or count variables and five categorical business attributes. Categorical variables were dummy
encoded with explicit reference categories, and an 80/20 stratified train/test split was used so model
performance could be evaluated on unseen observations.
Backward elimination removed four non-significant terms and retained nine predictors/dummy terms. At
a 0.50 classification threshold, the final test-set accuracy was 68.9%, recall was 54.5%, specificity was
79.6%, and ROC AUC was 0.734. The model therefore provides useful directional information and
moderate discrimination, but it should be treated as a decision-support model rather than an automated
churn decision engine. The strongest retained relationships are shorter tenure, higher monthly charges,
more support calls, month-to-month contracting, lack of automatic payment, and lack of online security.
Business conclusion
The model supports targeted retention outreach. Customers with short tenure, high charges, repeated
support contacts, month-to-month contracts, manual payment, and no online-security service should receive
proactive service recovery and contract-value offers. Model performance is adequate for prioritization, but
not strong enough to justify high-impact customer decisions without additional validation and richer
behavioral data.
1. Business Problem, Research Question, and Analytical Goal
1.1 Scenario
A telecommunications provider is experiencing preventable customer attrition. Leadership wants an
interpretable statistical model that identifies factors associated with churn and produces a customer-level
churn probability. The organization can use the results to prioritize retention resources, investigate
service pain points, and evaluate whether contract and account-management policies are associated with
customer loss.
1.2 Research question
Research question
To what extent do customer tenure, monthly charge, support-call frequency, contract type, internet service,
automatic payment, online security, paperless billing, age, and annual income predict the probability that a
telecommunications customer will churn?
1.3 Goal and practical objective
Estimate an interpretable multiple logistic regression model for a binary churn outcome.
Quantify the direction and size of each retained association using coefficients and odds ratios.
Reduce the initial model using a documented statistical criterion rather than subjective removal.
September 2026 aligned performance assessment model | Page 2
, WGU D600 - Statistical Data Mining | Task 2
Evaluate classification on unseen test data using a confusion matrix, accuracy, precision, recall,
specificity, F1 score, and ROC AUC.
Translate the results into retention actions while acknowledging statistical and data limitations.
1.4 Why logistic regression is appropriate
Logistic regression is appropriate because the dependent variable has two possible outcomes. Unlike
ordinary least squares, the model estimates the log-odds of churn and transforms the result into a
probability bounded between 0 and 1. The coefficients remain interpretable, and exponentiating a
coefficient yields an odds ratio that is useful for business communication.
2. Data, Variables, and Preparation
2.1 Dataset status
The dataset used here is a representative, reproducible telecommunications dataset constructed for this
model submission. It is not described as WGU proprietary data. The structure intentionally reflects a
realistic classification problem and contains enough variation to demonstrate the full D600 Task 2
workflow without fabricating access to a restricted learner-portal file.
2.2 Variables
Variable Type Role Operational definition
Churn Binary Dependent 1 = churned; 0 = retained.
Age Continuous Independent Customer age in years.
AnnualIncome Continuous Independent Estimated annual income in
USD.
TenureMonths Continuous Independent Months with the provider.
MonthlyCharge Continuous Independent Current monthly service
charge in USD.
SupportCalls Count Independent Recent customer-support
contacts.
Contract Nominal Independent Two-year, one-year, or month-
to-month.
InternetService Nominal Independent DSL, fiber, or no internet
service.
AutoPay Binary nominal Independent Automatic payment
enrollment.
OnlineSecurity Binary nominal Independent Online-security service
enrollment.
PaperlessBilling Binary nominal Independent Paperless billing enrollment.
2.3 Data-quality checks
Rows analyzed: 1,800; duplicate rows: 0; missing values: 0.
The dependent variable was confirmed to contain only 0 and 1.
September 2026 aligned performance assessment model | Page 3
WGU D600
Statistical Data Mining
TASK 2: MULTIPLE LOGISTIC REGRESSION ANALYSIS
Completed Performance Assessment Model | September 2026
Predicting Telecommunications Customer Churn with Multiple Logistic
Regression
Scope and current alignment
WGU identifies D600 as Statistical Data Mining in the current September 2026 catalog. The course focuses on
regression plus supervised classification and prediction. Publicly documented D600 workflow identifies Task
2 as logistic regression and describes version-controlled code, train/test evaluation, accuracy and confusion-
matrix evidence, model reduction, and a presentation. The complete live learner-portal rubric and assigned
dataset are not publicly available, so this document uses a representative telecommunications churn
scenario and explicitly avoids presenting invented material as proprietary WGU content.
Original academic model. It is not an official WGU exemplar and no evaluator-issued pass is claimed.
September 2026 aligned performance assessment model | Page 1
, WGU D600 - Statistical Data Mining | Task 2
Executive Summary
This analysis addresses a practical retention question: which customer attributes are associated with
churn, and how accurately can a multiple logistic regression model classify whether a customer will
leave? A representative dataset of 1,800 customer records was analyzed. The binary outcome Churn
equals 1 for a customer who leaves and 0 for a customer who remains. The initial model included five
continuous or count variables and five categorical business attributes. Categorical variables were dummy
encoded with explicit reference categories, and an 80/20 stratified train/test split was used so model
performance could be evaluated on unseen observations.
Backward elimination removed four non-significant terms and retained nine predictors/dummy terms. At
a 0.50 classification threshold, the final test-set accuracy was 68.9%, recall was 54.5%, specificity was
79.6%, and ROC AUC was 0.734. The model therefore provides useful directional information and
moderate discrimination, but it should be treated as a decision-support model rather than an automated
churn decision engine. The strongest retained relationships are shorter tenure, higher monthly charges,
more support calls, month-to-month contracting, lack of automatic payment, and lack of online security.
Business conclusion
The model supports targeted retention outreach. Customers with short tenure, high charges, repeated
support contacts, month-to-month contracts, manual payment, and no online-security service should receive
proactive service recovery and contract-value offers. Model performance is adequate for prioritization, but
not strong enough to justify high-impact customer decisions without additional validation and richer
behavioral data.
1. Business Problem, Research Question, and Analytical Goal
1.1 Scenario
A telecommunications provider is experiencing preventable customer attrition. Leadership wants an
interpretable statistical model that identifies factors associated with churn and produces a customer-level
churn probability. The organization can use the results to prioritize retention resources, investigate
service pain points, and evaluate whether contract and account-management policies are associated with
customer loss.
1.2 Research question
Research question
To what extent do customer tenure, monthly charge, support-call frequency, contract type, internet service,
automatic payment, online security, paperless billing, age, and annual income predict the probability that a
telecommunications customer will churn?
1.3 Goal and practical objective
Estimate an interpretable multiple logistic regression model for a binary churn outcome.
Quantify the direction and size of each retained association using coefficients and odds ratios.
Reduce the initial model using a documented statistical criterion rather than subjective removal.
September 2026 aligned performance assessment model | Page 2
, WGU D600 - Statistical Data Mining | Task 2
Evaluate classification on unseen test data using a confusion matrix, accuracy, precision, recall,
specificity, F1 score, and ROC AUC.
Translate the results into retention actions while acknowledging statistical and data limitations.
1.4 Why logistic regression is appropriate
Logistic regression is appropriate because the dependent variable has two possible outcomes. Unlike
ordinary least squares, the model estimates the log-odds of churn and transforms the result into a
probability bounded between 0 and 1. The coefficients remain interpretable, and exponentiating a
coefficient yields an odds ratio that is useful for business communication.
2. Data, Variables, and Preparation
2.1 Dataset status
The dataset used here is a representative, reproducible telecommunications dataset constructed for this
model submission. It is not described as WGU proprietary data. The structure intentionally reflects a
realistic classification problem and contains enough variation to demonstrate the full D600 Task 2
workflow without fabricating access to a restricted learner-portal file.
2.2 Variables
Variable Type Role Operational definition
Churn Binary Dependent 1 = churned; 0 = retained.
Age Continuous Independent Customer age in years.
AnnualIncome Continuous Independent Estimated annual income in
USD.
TenureMonths Continuous Independent Months with the provider.
MonthlyCharge Continuous Independent Current monthly service
charge in USD.
SupportCalls Count Independent Recent customer-support
contacts.
Contract Nominal Independent Two-year, one-year, or month-
to-month.
InternetService Nominal Independent DSL, fiber, or no internet
service.
AutoPay Binary nominal Independent Automatic payment
enrollment.
OnlineSecurity Binary nominal Independent Online-security service
enrollment.
PaperlessBilling Binary nominal Independent Paperless billing enrollment.
2.3 Data-quality checks
Rows analyzed: 1,800; duplicate rows: 0; missing values: 0.
The dependent variable was confirmed to contain only 0 and 1.
September 2026 aligned performance assessment model | Page 3
