MLD109 · Final Project: A Predictive Model for a Business Decision
About Course
Frame, build, evaluate, explain and package a predictive model that supports one real business decision, bringing together the full data science workflow.
Who it is for
Learners who have completed the MLD program courses and want a substantial, portfolio-ready project.
Prerequisites
MLD101 through MLD108, or equivalent experience across the data science workflow.
What you will be able to do
- Frame a business decision as a predictive problem with a target, metric, baseline and value estimate
- Prepare data and engineer features without leakage, using validation that mirrors real use
- Build, tune, explain and fairly evaluate a model against a baseline with a cost-based decision rule
- Package the model with a monitoring plan and present results, value and limitations in a decision-ready report
Lessons
- Lesson 1 · Project brief and scope
- Lesson 2 · Milestones and method
- Lesson 3 · Quality checklist and common mistakes
- Lesson 4 · Deliverables, rubric and submission
Lab: Project build environment
You will keep the entire project in one repository: notebooks or scripts, data documentation, MLflow runs, the packaged model, a monitoring report and the final report. Each milestone is committed so reviewers can follow your reasoning from framing to deployment.
Free tools: Google Colab or VS Code, GitHub free, MLflow (local), Evidently, Docker Desktop and Looker Studio (all free)
Assessed assignment
Choose one business decision, such as which customers to contact for renewal, how much stock to order per store, or which maintenance jobs to prioritize, using a public dataset or anonymized data you are allowed to use. Build a predictive model that supports the decision, compare it with a baseline under a validation scheme that matches real use, set a decision rule from costs and capacity, explain the model and check subgroup performance. Deliver the repository, a packaged model with a monitoring plan and a final report of 2,000 to 3,000 words with an executive summary.
| Criterion | Weight | What excellent work looks like |
|---|---|---|
| Problem framing and value | 20% | A specific decision, target, metric and baseline, with a realistic estimate of value and its assumptions. |
| Data and modeling rigor | 30% | Leak-free preparation, validation matching deployment, sound model comparison and tuning that never touches the test set. |
| Evaluation, explanation and fairness | 25% | Cost-based decision rule, test results with uncertainty, careful explanations and a subgroup analysis. |
| Deployment readiness and communication | 25% | A runnable packaged model, a monitoring plan and a clear report that a manager and a data professional can both use. |
Portfolio piece
A complete, portfolio-ready predictive modeling project with repository, packaged model, monitoring plan and decision report.
Next step
You have reached the final course of this program. Keep growing with a Learning Path that combines this program with related ones.
Course Content
MLD109 · Final Project: A Predictive Model for a Business Decision (8 h)
-
Lesson 1 · Project brief and scope
-
Lesson 2 · Milestones and method
-
Lesson 3 · Quality checklist and common mistakes
-
Lesson 4 · Deliverables, rubric and submission
-
MLD109 Assessment · Final Project: A Predictive Model for a Business Decision