WGU D606 EBN2 – Task 2: Data Analysis
Report
WGU D606 EBN2 – DATA SCIENCE
CAPSTONE
TASK 2: DATA ANALYSIS REPORT
Western Governors University
Program: Master of Science in Data Analytics – Data Science Specialization
Course: D606 – Data Science Capstone
Assessment: EBN2 Task 2 – Data Analysis Report
Academic Year: 2026/2027
Project Title: Predicting Mobile Phone Customer Satisfaction Using Multiple Linear
Regression: An Analysis of Product Features, Pricing, and Brand
Executive Summary
Understanding customer satisfaction is important to mobile phone manufacturers because
product development decisions require substantial financial resources and must respond to
changing consumer expectations. Customer review datasets provide opportunities to investigate
how evaluations of individual product features relate to overall satisfaction.
This study examines the predictive relationship between overall mobile phone customer ratings
and seven groups of explanatory variables: brand, price, battery life, camera quality,
performance, physical design, and display quality.
The analysis draws on the Global Mobile Reviews Dataset (2025 Edition), described in the
accompanying research documentation as containing 50,000 customer reviews collected across
eight countries between October 2022 and October 2025.
Python was used in the documented analytical workflow for data preparation, descriptive
statistics, exploratory analysis, multiple linear regression, feature selection, and model
evaluation.
The reported analysis used an 80% training and 20% testing split. An initial ordinary least
squares regression incorporated twelve encoded predictors. Backward elimination produced a
reduced model containing five feature-specific rating variables.
,The reduced model achieved a reported training R-squared of 0.863 and a test-set R-squared of
0.8657. Mean absolute error was 0.3691 rating points, while root mean squared error was 0.4555
points.
The results indicate that feature-specific ratings contain substantial predictive information about
overall customer ratings within the reported dataset. Brand and price did not demonstrate
statistically significant additional contributions in the specified full regression.
However, interpretation must account for the observational design, potential dependence
between overall and feature-specific ratings, correlated explanatory variables, and uncertainty
about the original data-generation process.
The study concludes that multiple linear regression provides a useful model for estimating
overall ratings in this dataset. Its findings are most appropriately applied to customer feedback
monitoring and further product-quality investigation rather than direct causal claims about
product development priorities.
A. RESEARCH QUESTION
A1. Research Question
The primary research question is:
Can a multiple linear regression model be constructed using mobile phone feature-specific
ratings, brand, and price to predict overall customer satisfaction with a test-set R-squared
greater than 0.70?
The analysis examines whether numerical evaluations of important device characteristics,
combined with product pricing and manufacturer information, can explain a substantial
proportion of the variation in customer satisfaction ratings.
The model uses overall customer rating as the dependent variable and product-related
information as explanatory variables.
The analysis also evaluates whether removing statistically nonsignificant predictors produces a
simpler model without materially reducing predictive performance.
A2. Justification of the Research Question
Mobile phone manufacturers make decisions about product design, technical specifications,
quality improvement, and research and development investments in competitive markets.
,Customer satisfaction is influenced by various factors, including battery life, camera quality,
processing performance, screen quality, product design, and perceived value.
Manufacturers must determine how to interpret these factors when developing future products.
A customer may assign a high rating to a device because several important features perform well.
Conversely, weaknesses in particular features may contribute to dissatisfaction.
However, individual feature ratings may be related to one another, making it difficult to
determine their independent associations with overall satisfaction.
Multiple linear regression provides a statistical framework for examining several predictors
simultaneously.
The technique estimates an overall linear relationship between explanatory variables and the
response variable while controlling for the other included predictors.
This is particularly useful when evaluating whether brand and price provide additional predictive
value beyond customers' evaluations of functional product characteristics.
Research examining mobile phone reviews has identified battery performance, camera features,
display characteristics, price, and device functionality as important topics in customer feedback
(Aggarwal et al., 2025).
The present analysis extends that general research context by examining structured numerical
feature ratings and comparing the predictive performance of full and reduced regression models.
The findings can support more systematic interpretation of customer satisfaction data and
establish a basis for subsequent product-quality research.
A3. Business Context
Mobile phone manufacturers continuously introduce devices with different specifications, prices,
and intended customer segments.
Product managers must balance technical improvements with manufacturing costs, development
time, and customer expectations.
Although manufacturers often collect customer ratings and reviews, descriptive summaries alone
may not reveal how different product evaluations jointly relate to overall satisfaction.
A predictive model can help organizations identify patterns in customer feedback and understand
whether particular rating dimensions carry useful information.
The business problem addressed by this project is the need for a transparent, reproducible
method of predicting overall customer satisfaction from structured product evaluations.
, The model is intended to support research and development analysis, quality monitoring, and
customer feedback interpretation.
The analysis does not establish that making a particular product change will cause customer
satisfaction to improve. Such claims would require additional causal evidence or controlled
testing.
A4. Research Hypotheses
The project establishes a predictive performance benchmark of R-squared greater than 0.70.
Null Hypothesis (H₀):
The proposed regression model will not achieve a test-set R-squared greater than 0.70.
H₀: Test-set R² ≤ 0.70
Alternative Hypothesis (H₁):
The proposed regression model will achieve a test-set R-squared greater than 0.70.
H₁: Test-set R² > 0.70
These hypotheses express a predictive performance objective. A single observed test-set R-
squared above 0.70 demonstrates that the reported benchmark was achieved, but it does not by
itself constitute a formal significance test of the population-level R-squared.
For conventional regression inference, a separate overall model hypothesis is also considered:
H₀: β₁ = β₂ = ... = βₖ = 0
H₁: At least one βⱼ differs from zero.
The overall F-test is used to assess whether the estimated regression provides evidence of a
linear association beyond an intercept-only model.
A significance level of 0.05 is applied to conventional inferential testing.
A5. Research Objectives
The analytical objectives are to:
1. Prepare the mobile phone review dataset for statistical analysis.
2. Examine descriptive distributions and correlations among the selected variables.
3. Construct an initial multiple linear regression model.
Report
WGU D606 EBN2 – DATA SCIENCE
CAPSTONE
TASK 2: DATA ANALYSIS REPORT
Western Governors University
Program: Master of Science in Data Analytics – Data Science Specialization
Course: D606 – Data Science Capstone
Assessment: EBN2 Task 2 – Data Analysis Report
Academic Year: 2026/2027
Project Title: Predicting Mobile Phone Customer Satisfaction Using Multiple Linear
Regression: An Analysis of Product Features, Pricing, and Brand
Executive Summary
Understanding customer satisfaction is important to mobile phone manufacturers because
product development decisions require substantial financial resources and must respond to
changing consumer expectations. Customer review datasets provide opportunities to investigate
how evaluations of individual product features relate to overall satisfaction.
This study examines the predictive relationship between overall mobile phone customer ratings
and seven groups of explanatory variables: brand, price, battery life, camera quality,
performance, physical design, and display quality.
The analysis draws on the Global Mobile Reviews Dataset (2025 Edition), described in the
accompanying research documentation as containing 50,000 customer reviews collected across
eight countries between October 2022 and October 2025.
Python was used in the documented analytical workflow for data preparation, descriptive
statistics, exploratory analysis, multiple linear regression, feature selection, and model
evaluation.
The reported analysis used an 80% training and 20% testing split. An initial ordinary least
squares regression incorporated twelve encoded predictors. Backward elimination produced a
reduced model containing five feature-specific rating variables.
,The reduced model achieved a reported training R-squared of 0.863 and a test-set R-squared of
0.8657. Mean absolute error was 0.3691 rating points, while root mean squared error was 0.4555
points.
The results indicate that feature-specific ratings contain substantial predictive information about
overall customer ratings within the reported dataset. Brand and price did not demonstrate
statistically significant additional contributions in the specified full regression.
However, interpretation must account for the observational design, potential dependence
between overall and feature-specific ratings, correlated explanatory variables, and uncertainty
about the original data-generation process.
The study concludes that multiple linear regression provides a useful model for estimating
overall ratings in this dataset. Its findings are most appropriately applied to customer feedback
monitoring and further product-quality investigation rather than direct causal claims about
product development priorities.
A. RESEARCH QUESTION
A1. Research Question
The primary research question is:
Can a multiple linear regression model be constructed using mobile phone feature-specific
ratings, brand, and price to predict overall customer satisfaction with a test-set R-squared
greater than 0.70?
The analysis examines whether numerical evaluations of important device characteristics,
combined with product pricing and manufacturer information, can explain a substantial
proportion of the variation in customer satisfaction ratings.
The model uses overall customer rating as the dependent variable and product-related
information as explanatory variables.
The analysis also evaluates whether removing statistically nonsignificant predictors produces a
simpler model without materially reducing predictive performance.
A2. Justification of the Research Question
Mobile phone manufacturers make decisions about product design, technical specifications,
quality improvement, and research and development investments in competitive markets.
,Customer satisfaction is influenced by various factors, including battery life, camera quality,
processing performance, screen quality, product design, and perceived value.
Manufacturers must determine how to interpret these factors when developing future products.
A customer may assign a high rating to a device because several important features perform well.
Conversely, weaknesses in particular features may contribute to dissatisfaction.
However, individual feature ratings may be related to one another, making it difficult to
determine their independent associations with overall satisfaction.
Multiple linear regression provides a statistical framework for examining several predictors
simultaneously.
The technique estimates an overall linear relationship between explanatory variables and the
response variable while controlling for the other included predictors.
This is particularly useful when evaluating whether brand and price provide additional predictive
value beyond customers' evaluations of functional product characteristics.
Research examining mobile phone reviews has identified battery performance, camera features,
display characteristics, price, and device functionality as important topics in customer feedback
(Aggarwal et al., 2025).
The present analysis extends that general research context by examining structured numerical
feature ratings and comparing the predictive performance of full and reduced regression models.
The findings can support more systematic interpretation of customer satisfaction data and
establish a basis for subsequent product-quality research.
A3. Business Context
Mobile phone manufacturers continuously introduce devices with different specifications, prices,
and intended customer segments.
Product managers must balance technical improvements with manufacturing costs, development
time, and customer expectations.
Although manufacturers often collect customer ratings and reviews, descriptive summaries alone
may not reveal how different product evaluations jointly relate to overall satisfaction.
A predictive model can help organizations identify patterns in customer feedback and understand
whether particular rating dimensions carry useful information.
The business problem addressed by this project is the need for a transparent, reproducible
method of predicting overall customer satisfaction from structured product evaluations.
, The model is intended to support research and development analysis, quality monitoring, and
customer feedback interpretation.
The analysis does not establish that making a particular product change will cause customer
satisfaction to improve. Such claims would require additional causal evidence or controlled
testing.
A4. Research Hypotheses
The project establishes a predictive performance benchmark of R-squared greater than 0.70.
Null Hypothesis (H₀):
The proposed regression model will not achieve a test-set R-squared greater than 0.70.
H₀: Test-set R² ≤ 0.70
Alternative Hypothesis (H₁):
The proposed regression model will achieve a test-set R-squared greater than 0.70.
H₁: Test-set R² > 0.70
These hypotheses express a predictive performance objective. A single observed test-set R-
squared above 0.70 demonstrates that the reported benchmark was achieved, but it does not by
itself constitute a formal significance test of the population-level R-squared.
For conventional regression inference, a separate overall model hypothesis is also considered:
H₀: β₁ = β₂ = ... = βₖ = 0
H₁: At least one βⱼ differs from zero.
The overall F-test is used to assess whether the estimated regression provides evidence of a
linear association beyond an intercept-only model.
A significance level of 0.05 is applied to conventional inferential testing.
A5. Research Objectives
The analytical objectives are to:
1. Prepare the mobile phone review dataset for statistical analysis.
2. Examine descriptive distributions and correlations among the selected variables.
3. Construct an initial multiple linear regression model.
