🚧 Under construction — VoTechWorks is being built. A free tech program — complimentary, as our commitment to get America working.
voTechWorksGrade 13 · a vocational technology learning center, by urbanicity
newSchoolWorks.com › VoTechWorks › Technology & Design › Data Science & Analytics › Capstone Project: End-to-End Data Science & Analytics Build › Lesson 3
TEC-DATA85-12 · Lesson 3 · 72 clock hours · lab

Modeling, Dashboard Build, and Report Drafting

Objective. Students train and evaluate at least one machine learning model appropriate to their question, build a fully interactive Power BI dashboard presenting key findings, and produce a complete draft project report that integrates all prior work.

⚠ Pending professional review

This lesson, its worked examples and its practice questions has not yet been verified by an industry-certified professional or by a certification or approval body. It is a study draft and may contain errors. Do not use it as instructions for work on a job site or on live equipment. Always follow your instructor, your supervising professional, and current official references and regulations.

VoTechWorks program approval and accreditation are still in progress. Credentials are issued only by the certifying bodies named in each program, after their own exams.

Modeling, Dashboard Build, and Report Drafting

This lab brings together every skill you have built across the program. By the end of this session you will have a trained and evaluated machine learning model, a live interactive Power BI dashboard, and a complete draft project report ready for instructor review. Each of these three deliverables maps directly to credential competency areas, so quality here has real exam value.

---

Step 1: Choosing and Fitting Your Model

Before writing a single line of code, write a half-page model-selection memo. Explain which modeling approach you chose—classification, regression, clustering, or time-series forecasting—and why it matches your research question and target variable. A classification model predicts a category (e.g., "will this customer churn?"). A regression model predicts a continuous value (e.g., "what will sales be next quarter?"). Clustering finds natural groupings when you have no labeled outcome. Time-series forecasting handles data ordered over time.

Once justified, split your cleaned dataset into a training set and a test set (a common split is 80 % training, 20 % test, though the right ratio depends on dataset size). Never let the model see test data during training—this is how you get an honest estimate of real-world performance.

Fit your model using scikit-learn (or statsmodels for statistical approaches). Then compute at least two appropriate performance metrics:

Document every hyperparameter you set and any cross-validation steps (e.g., k-fold CV) inside your notebook. This documentation is a core expectation in the IBM Data Science Professional Certificate ML competency areas.

---

Step 2: Writing Metric Interpretations

A number alone is not an insight. Every metric must be followed by a plain-English sentence that ties it back to the business question. For example: "The model achieved an RMSE of 4,200 units, meaning predictions are off by roughly 4,200 units on average—acceptable given that the planning team works in 10,000-unit batches." That kind of interpretation is what distinguishes a data professional from someone who just runs code.

---

Step 3: Building the Power BI Dashboard

Open Power BI Desktop (free download from Microsoft) and connect it to your processed dataset. Your finished .pbix file must contain:

Use DAX (Data Analysis Expressions) to create at least one calculated measure rather than relying only on automatic aggregations. A DAX measure makes your model flexible and is a tested skill in the Microsoft PL-300 exam domains covering data modeling and report design.

When the report is complete, publish it to Power BI Service and generate a shareable live link. Verify the link opens correctly in a private browser window before submitting it. A broken link at the defense is a common, avoidable error.

---

Step 4: Drafting the Project Report

Your report follows standard technical report structure. Every section must be present before the defense is scheduled:

1. Executive Summary — one page maximum; key findings and recommendation.

2. Problem Statement — what question are you answering, and why does it matter?

3. Data Description and Ethics Review — source, size, variables, any privacy or bias considerations.

4. Methodology — EDA steps, cleaning decisions, modeling approach, and justification.

5. Results — EDA findings, model metrics with written interpretations.

6. Dashboard Description — explain each visual and what it communicates.

7. Limitations — what could affect the validity or generalizability of your findings?

8. Recommendations — what should a decision-maker do based on your analysis?

9. References — APA format for all datasets, packages, and sources cited.

Submit the draft to your instructor using the written report rubric. Instructor feedback is part of the process; incorporate it before the final defense.

---

Data Ethics and Safety Reminders

Even in a software-only environment, professional standards apply. Before publishing any dashboard or report, verify that no personally identifiable information (PII) appears in visuals or exported data. If your dataset required an ethics review, confirm that your published outputs comply with the data-use agreement you documented earlier in the project. In professional practice, publishing a live link with sensitive data exposed is a serious compliance failure. Treat data security with the same discipline a technician applies to physical safety procedures on the shop floor.

Key terms9

Training/test splitDividing a dataset into a portion used to fit the model (training set) and a separate portion used to evaluate its performance on unseen data (test set), preventing overly optimistic accuracy estimates.
F1-scoreA classification metric that combines precision and recall into a single value using their harmonic mean; useful when class sizes are unequal and both false positives and false negatives matter.
RMSE (Root Mean Squared Error)A regression metric that measures the average magnitude of prediction errors in the same units as the target variable; larger errors are penalized more heavily than smaller ones because errors are squared before averaging.
R² (Coefficient of Determination)A regression metric expressing the proportion of variance in the target variable that the model explains; a value of 1.0 indicates a perfect fit, while 0.0 means the model explains none of the variance.
DAX (Data Analysis Expressions)A formula language used in Power BI to create calculated columns, measures, and tables; tested in the Microsoft PL-300 exam and essential for building flexible, reusable report logic.
Cross-validationA model evaluation technique—commonly k-fold—that repeatedly splits the training data into different training and validation subsets to produce a more reliable estimate of model performance than a single split.
KPI (Key Performance Indicator)A measurable value that indicates progress toward a defined business goal; in Power BI, KPIs are often displayed on card visuals so decision-makers can see the headline figure immediately.
HyperparameterA setting configured before model training (e.g., the number of trees in a random forest or the regularization strength in a regression) that controls learning behavior and must be documented and justified.
Executive summaryA concise opening section of a technical report—typically one page or less—that states the problem, key findings, and recommended action so a busy stakeholder can grasp the project without reading the full document.

Worked examples4

1. A student trains a logistic regression classifier to predict customer churn. On the test set of 500 records, the model correctly labels 420 records. Of the 80 records the model predicts as 'will churn,' 60 actually churned. Of all 130 true churners in the test set, the model correctly identified 60. Compute accuracy and F1-score, then interpret both for the business.
  1. Accuracy = correct predictions ÷ total records = 420 ÷ 500 = 0.84 (84 %).
  2. Precision = true positives ÷ predicted positives = 60 ÷ 80 = 0.75.
  3. Recall = true positives ÷ actual positives = 60 ÷ 130 ≈ 0.462.
  4. F1-score = 2 × (Precision × Recall) ÷ (Precision + Recall) = 2 × (0.75 × 0.462) ÷ (0.75 + 0.462) = 2 × 0.3465 ÷ 1.212 ≈ 0.572.
  5. Write the business interpretation: 'The model is correct 84 % of the time overall, but the F1-score of 0.57 reveals it misses roughly half of all true churners—a meaningful risk if retention campaigns depend on this model to target the right customers.'
Answer: Accuracy = 0.84; F1-score ≈ 0.57. The gap between the two metrics signals that the model struggles to recall true churners, which is the more business-critical class.
2. A regression model predicts monthly electricity usage (kWh). On a test set of 10 homes, the squared errors between predicted and actual kWh are: 900, 400, 1600, 100, 2500, 625, 225, 1024, 784, 441. Compute RMSE.
  1. Sum all squared errors: 900 + 400 + 1600 + 100 + 2500 + 625 + 225 + 1024 + 784 + 441 = 8599.
  2. Divide by the number of observations to get Mean Squared Error (MSE): 8599 ÷ 10 = 859.9.
  3. Take the square root: RMSE = √859.9 ≈ 29.3 kWh.
  4. Interpret: 'On average, the model's predictions are off by about 29 kWh per month, which the operations team must weigh against typical monthly usage levels when deciding whether the model is accurate enough for planning.'
Answer: RMSE ≈ 29.3 kWh.
3. A student needs a DAX measure in Power BI that calculates the percentage of orders that arrived late. The table 'Orders' has columns 'TotalOrders' (count) and 'LateOrders' (count). Write and explain the DAX measure.
  1. In Power BI Desktop, go to the Modeling tab and select 'New Measure.'
  2. Enter the formula: Late Order Rate = DIVIDE(SUM(Orders[LateOrders]), SUM(Orders[TotalOrders]), 0)
  3. DIVIDE is preferred over the division operator ( / ) because it safely returns 0 (the third argument) when the denominator is zero, preventing errors.
  4. Format the measure as a percentage in the Measure tools ribbon.
  5. Place the measure on a Card visual and add it to a slicer for Region so users can see the late-order rate for any region they select.
Answer: Late Order Rate = DIVIDE(SUM(Orders[LateOrders]), SUM(Orders[TotalOrders]), 0) — a robust measure that returns the proportion of late orders and handles zero denominators safely.
4. A student uses 5-fold cross-validation in scikit-learn on a training set of 1,000 records. The five validation fold accuracy scores are 0.81, 0.84, 0.79, 0.83, and 0.83. Compute the mean CV accuracy and standard deviation, then explain what these values say about model stability.
  1. Mean = (0.81 + 0.84 + 0.79 + 0.83 + 0.83) ÷ 5 = 4.10 ÷ 5 = 0.82.
  2. Deviations from mean: −0.01, +0.02, −0.03, +0.01, +0.01.
  3. Squared deviations: 0.0001, 0.0004, 0.0009, 0.0001, 0.0001. Sum = 0.0016.
  4. Variance = 0.0016 ÷ 5 = 0.00032. Standard deviation = √0.00032 ≈ 0.0179, rounded to 0.018.
  5. Interpret: 'The mean CV accuracy of 0.82 ± 0.018 suggests the model performs consistently across different subsets of the training data, which reduces the risk that good test-set results are due to a lucky split.'
Answer: Mean CV accuracy = 0.82; standard deviation ≈ 0.018. Low standard deviation indicates the model is stable and not overly sensitive to which records land in which fold.

The math in this lesson1

Learners compute accuracy, precision, recall and F1 from model prediction counts.
S-CP.4
Brush up in the math halls at newSchoolWorks.com:
In class: Students select a modeling approach (classification, regression, clustering, or time-series forecasting) justified in a half-page model-selection memo; they split data into training and test sets, fit the model using scikit-learn or equivalent, compute at least two appropriate performance metrics (e.g., accuracy and F1 for classification; RMSE and R² for regression), and document hyperparameter choices and any cross-validation steps in their notebook—directly aligned to IBM Data Science Professional Course 6–9 (ML with Python) competency areas. · Students build a Power BI report file (.pbix) containing at least four visuals (including one map visual if geographic data is present), a slicer for at least one dimension, a card showing a key KPI, and an executive-summary text box; they publish to Power BI Service and share a live link; the dashboard is evaluated against the Microsoft PL-300 domain areas covering data modeling, report design, and DAX measures. · Students draft the complete project report following a standard technical report structure: executive summary, problem statement, data description and ethics review, methodology, results (EDA + model), dashboard description, limitations, recommendations, and references in APA format; draft is submitted for instructor feedback using a written report rubric before the defense.
Materials: Python with scikit-learn, statsmodels; Power BI Desktop (free) and Power BI Service account; DAX reference card; Microsoft PL-300 exam domain guide; IBM Data Science Professional Certificate ML competency checklist; project report template in .docx or Google Docs; APA citation guide; screen-recording software (OBS or equivalent) for dashboard walkthrough video.
Check: Model evaluation section must include at least two correctly computed metrics with a written interpretation of what the numbers mean for the business question. Power BI dashboard must be live and shareable; evaluated on a 15-point PL-300-aligned rubric—minimum 11/15. Complete report draft must be submitted by the scheduled deadline; instructor must confirm all required sections are present before the defense is scheduled.

Practice check8

Answer each question, then check it. Nothing is recorded; this is your own practice.

IBM Data Science Professional — Machine Learning with Python
1. A student's binary classification model achieves 95 % accuracy on a test set where 95 % of records belong to the negative class. Which statement best describes this result?
IBM Data Science Professional — Machine Learning with Python
2. Which scikit-learn function is the standard way to divide a dataset into training and test portions while shuffling the data randomly?
Microsoft PL-300 — Create Measures Using DAX
3. In the context of the Microsoft PL-300 exam, what is the primary advantage of using the DAX DIVIDE function instead of the forward-slash operator ( / ) when creating a calculated measure?
Microsoft PL-300 — Design and Build Reports
4. A Power BI report must display sales figures filterable by product category and by year. Which visual element should the student add to enable users to interactively filter all other visuals on the page?
IBM Data Science Professional — Machine Learning with Python
5. RMSE and R² are both computed for a regression model. RMSE is 120 and R² is 0.91. Which interpretation is correct?
Google Data Analytics Certificate — Share Data Through the Art of Visualization
6. Which section of a standard technical project report is designed to give a busy executive a complete understanding of the project without reading the full document?
CompTIA Data+ — Data Governance, Quality, and Controls
7. Before publishing a Power BI dashboard with a live share link, a student working with customer data should first:
IBM Data Science Professional — Machine Learning with Python
8. A student runs 5-fold cross-validation and receives accuracy scores of 0.70, 0.72, 0.69, 0.71, and 0.68. The mean is approximately 0.70. What does the low standard deviation across folds indicate?
Record this lesson on my transcript. Sign in with your key and enrollment email to record lessons (enroll free if you have no key). This page stays open to read and practise.
← Previous lessonProgram planNext lesson →