• Wrong document? Swap it for free
  • Written by students who passed
  • Immediately available after payment
  • Read online or as PDF
Sell
Where do you study
Your language
Document preview thumbnail
Preview 3 out of 16 pages
Other

WGU D682 Task 3 – Artificial Intelligence Optimization for Computer Scientists | Data Assumptions, Model Interpretation & Predictive Analysis | 2026/2027 Updated

Document preview thumbnail
Preview 3 out of 16 pages

WGU D682 Task 3 – Artificial Intelligence Optimization for Computer Scientists | Data Assumptions, Model Interpretation & Predictive Analysis | 2026/2027 Updated

Content preview

WGU D682 - Task 3
Artificial Intelligence Optimization for Computer
Scientists

Data Assumptions, Model Interpretation, and Predictive Analysis


D682 - Artificial Intelligence Optimization for
Course
Computer Scientists
Task 3 - Data Assumptions, Interpretation, and
Assessment
Prediction
Gradient Boosting regression model selected in
Model carried forward
Task 2
Version 2026/2027 reference edition


Practice-data disclosure: This report uses the same complete 200-row synthetic practice dataset created for
the companion Task 1 and Task 2 reference documents. It is not WGU-supplied data and is not represented as
an official WGU workbook. All reported metrics and predictions in this PDF are reproducible from that
practice dataset.



Prepared as a professionally structured, rubric-aligned reference for the D682 Task 3 competency.
Students should compare the organization and terminology with the current assessment page in their
own WGU course environment.




WGU D682 Task 3 Reference | Practice Dataset | Page 1

,Executive Summary
Task 3 evaluates whether the optimized artificial-intelligence solution can be interpreted responsibly
and used for prediction rather than merely reporting that an optimization improved a score. The
optimized Gradient Boosting regression model carried forward from Task 2 was evaluated on the
same fixed 40-row holdout set. It achieved RMSE 2.0358, MAE 1.6126, and R-squared 0.8878. Those
results indicate strong predictive fit on the synthetic practice data, but they do not by themselves
establish that the model would generalize to a different population, location, time period, or
operational environment.
The analysis below makes the assumptions explicit, interprets feature influence and residual
behavior, demonstrates predictions for new illustrative observations, and identifies the limitations
that would have to be addressed before a comparable model could support an operational
environmental-health decision. This distinction between measured test performance and justified
real-world use is central to responsible predictive analytics.
Measure Task 3 value Interpretation
Typical error magnitude with larger
RMSE 2.0358
misses weighted more heavily.
Average absolute difference between
MAE 1.6126
observed and predicted scores.
Approximately 88.8% of holdout
R-squared 0.8878 variance explained in this practice
sample.
Average signed error; a value near
Mean residual (bias check) 0.6643 zero suggests little overall directional
bias on the holdout set.
About 90% of holdout absolute errors
were at or below this value; used only
90th percentile absolute residual 2.8802
as a descriptive practice uncertainty
band.




1. Assessment Context and Continuity Across Tasks
WGU describes D682 as a course that moves from implementation and testing to model optimization,
then to creating data assumptions and interpretations for predictive analytics and future data
forecasting, and finally to adapting the AI solution to additional scenarios. Task 3 therefore sits
between technical optimization and later model adaptation. Its purpose is not to produce another
competing model; it is to show that the selected optimized model can be interpreted, bounded by
explicit assumptions, and used to generate defensible predictions.
Current student-completion evidence also indicates that Task 3 is normally submitted as a single
narrative PDF rather than a new GitLab implementation. For that reason, this reference report
carries forward the Task 2 model and its fixed test set. No new optimization was performed for Task
3, avoiding post-hoc tuning to the holdout data.

2. Practice Dataset and Model Carried Forward
The practice dataset contains 200 fully populated records, ten numeric predictors, and one
continuous target named healthRiskScore. The first 160 observations in the deterministic train/test
split are used for model fitting and 40 observations are held out for evaluation. The split uses
WGU D682 Task 3 Reference | Practice Dataset | Page 2

, random_state=42, exactly as in Tasks 1 and 2, so the performance comparisons remain directly
traceable.
Variable Role Practice interpretation
Ambient temperature in degrees
temperatureF Predictor
Fahrenheit.
humidityPct Predictor Relative humidity percentage.
windSpeedMph Predictor Wind speed in miles per hour.
pressureHpa Predictor Atmospheric pressure in hectopascals.
precipitationIn Predictor Precipitation amount in inches.
visibilityMiles Predictor Visibility distance in miles.
uvIndex Predictor UV-index-like practice indicator.
pollenIndex Predictor Practice pollen-burden indicator.
Practice community asthma prevalence
asthmaPrevalencePct Predictor
percentage.
populationDensity Predictor Practice population-density measure.
Continuous practice health-risk score
healthRiskScore Target
used for regression.




The dataset is synthetic and was created only to make this reference analysis complete and
reproducible. The variables are plausible for an environmental-health modeling exercise, but they
must not be confused with the hidden WGU workbook. In an actual submission, the student should
replace the practice definitions with the official feature-description sheet and report any mismatch
between the case-study narrative and the supplied data.

3. Data Assumptions
3.1 Assumption: the selected variables are measured consistently
The first assumption is that each predictor uses a stable definition, scale, and measurement process
across the records used for training, evaluation, and prediction. A temperature value must mean the
same thing from one record to another; population density must be computed on a comparable
geographic basis; and a score such as pollenIndex must retain the same scale. If measurement
definitions change, the model may interpret a numerical shift as a real environmental change when
it is actually a data-collection change.

3.2 Assumption: the observations are sufficiently representative of intended use
The second assumption is that the training observations represent the conditions in which
predictions will later be requested. NIST guidance emphasizes that test data and evaluation
conditions should be representative of expected use and that limitations on generalization should be
documented. This matters because a model can perform well on a familiar holdout sample and still
fail in a different region, season, population, or sensor configuration. The synthetic practice data
cannot establish external validity; it can demonstrate the analysis process only.

3.3 Assumption: healthRiskScore is a valid regression target
This analysis assumes that healthRiskScore is a meaningful continuous outcome and that differences
of several score units have a consistent interpretation across the scale. That makes regression metrics
such as RMSE and MAE appropriate. In an operational project, the organization would need to
document how the target was constructed, who validated it, and whether it is sufficiently connected


WGU D682 Task 3 Reference | Practice Dataset | Page 3

Document information

Uploaded on
September 28, 2026
Number of pages
16
Written in
2026/2027
Type
Other
Person
Unknown
$17.99

Wrong document? Swap it for free Within 14 days of purchase and before downloading, you can choose a different document. You can simply spend the amount again.
Written by students who passed
Immediately available after payment
Read online or as PDF

Seller avatar
Reputation scores are based on the amount of documents a seller has sold for a fee and the reviews they have received for those documents. There are three levels: Bronze, Silver and Gold. The better the reputation, the more your can rely on the quality of the sellers work.
Wiseman
3.9
(1702)
Sold
8776
Followers
3921
Items
32299
Last sold
3 hours ago




Why students choose Stuvia

Created by fellow students, verified by reviews

Quality you can trust: written by students who passed their exams and reviewed by others who've used these notes.

Didn't get what you expected? Choose another document

No worries! You can immediately select a different document that better matches what you need.

Pay how you prefer, start learning right away

No subscription, no commitments. Pay the way you're used to via credit card or EFT and download your PDF document instantly.

Student with book image

“Bought, downloaded, and aced it. It really can be that simple.”

Alisha Student

Working on your references?

Create accurate citations in APA, MLA and Harvard with our free citation generator.

Working on your references?

Frequently asked questions