Pandas AI Learning Platform Analytics Capstone

Overview

This capstone turns Pandas from a collection of methods into one complete analytics workflow. You will work as the data owner for a fictional AI learning platform that teaches Python, Pandas, analytics, and AI skills. The product team wants to know which learners are succeeding, which courses need improvement, how recommendation models perform, and what learner-level features could support future AI experiments.

The project is intentionally designed as a portfolio project. By the end, you should be able to explain the business problem, defend your cleaning choices, validate joins, build reports at the right grain, and export a model-ready learner feature table without leaking future information into training data.

The data is original synthetic data created for this project. It is not copied from a course notebook, public dataset, or third-party platform.

What You Will Build

• A data quality audit that catches missing values, duplicate natural events, invalid scores, outlier study sessions, and unresolved tickets.
• A cleaned relational Pandas workspace with learners, courses, lessons, enrollments, events, assessments, feedback, certificates, support tickets, recommendations, and content plans.
• A course performance report with enrollment, revenue, engagement, ratings, certificates, and recommendation metrics.
• A learner feature table for Data Science and AI use cases.
• Weekly time-series reports with resampling, rolling averages, lag features, and percent change.
• Text feedback summaries using vectorized string methods and token-level analysis.
• Recommendation performance reports by model version, course, category, rank position, click-through rate, and enrollment conversion.
• A final portfolio summary that explains practical business actions.

Dataset Guide

The project includes 12 CSV files. Each file is small enough to inspect manually but connected enough to practice real Pandas workflows.

• pandas_capstone_ai_recommendations.csv: 420 rows, 11 columns
• pandas_capstone_assessments.csv: 320 rows, 9 columns
• pandas_capstone_certificates.csv: 90 rows, 7 columns
• pandas_capstone_content_calendar.csv: 80 rows, 10 columns
• pandas_capstone_courses.csv: 14 rows, 9 columns
• pandas_capstone_data_dictionary.csv: 102 rows, 5 columns
• pandas_capstone_enrollments.csv: 260 rows, 9 columns
• pandas_capstone_feedback.csv: 260 rows, 9 columns
• pandas_capstone_learners.csv: 120 rows, 11 columns
• pandas_capstone_learning_events.csv: 900 rows, 10 columns
• pandas_capstone_lessons.csv: 56 rows, 7 columns
• pandas_capstone_support_tickets.csv: 150 rows, 10 columns

Important relationships:

• learners.learner_id joins to enrollments, events, assessments, feedback, tickets, and recommendations.
• courses.course_id joins to lessons, enrollments, events, assessments, feedback, certificates, and content calendar.
• lessons.lesson_id joins to learning events.
• enrollments.enrollment_id joins to events, assessments, feedback, certificates, and support tickets.
• recommendations.recommended_course_id joins back to courses.course_id.

Intentional issues to practice:

• Missing country, rating, coupon_code, feedback_text, and resolved_at values.
• Whitespace and casing inconsistencies in country, plan, and category.
• Outliers in minutes_spent, score_pct, price_paid_usd, and response_hours.
• Duplicate natural learning events with unique technical event IDs.
• Completed enrollments without certificates.
• Recommendation clicks that do not convert into enrollments.

Portfolio Submission Checklist

• Your repository has a clear README with the business problem, data model, and final outputs.
• Your code can run from a clean checkout.
• Your cleaning choices are documented.
• Your joins include validation checks.
• Your exported reports are stored in outputs/.
• Your learner feature table has one row per learner.
• Your final write-up includes three business recommendations.
• Your leakage notes explain which columns are safe for modeling and which are not.

FAQ

Is this project enough to practice Pandas for Data Science and AI?

Yes. It covers the Pandas workflows that show up repeatedly in DS and AI projects: ingestion, inspection, cleaning, joins, aggregation, reshaping, datetime analysis, text preprocessing, feature engineering, validation, and export.

Why is this not a machine learning project?

This capstone prepares the data that a machine learning project would need. Model training is intentionally out of scope so the focus stays on Pandas depth and data quality.

Can I use notebooks instead of scripts?

Yes. A notebook is fine for exploration. For portfolio quality, keep a script version of the final pipeline so another person can rerun it.

Why are some values messy on purpose?

Real datasets are rarely clean. The missing values, casing issues, outliers, and duplicate natural events force you to practice judgment instead of only memorizing syntax.

What should premium users compare against?

Compare your structure, outputs, join validation, feature grain, and leakage notes against the reference solution. The exact metric values may differ if your outlier policy differs.