Business Context

lululemon wants to improve conversion from digital class browsing to completed booking in lululemon Studio. The analytics team needs a predictive model that scores each browse session so growth and CRM teams can prioritize high-intent users and identify the strongest drivers of booking.

Dataset

You are given a session-level dataset built from lululemon Studio app and web events over the last 9 months.

• Size: 420K browse sessions, 44 features
• Target: Binary — booked a class within 24 hours of the browse session
• Class balance: 18% positive, 82% negative
• Missing data: ~12% missing in affinity scores for new users, ~6% missing in commerce fields for users with no purchase history

Success Criteria

A good solution should outperform a logistic regression baseline and achieve enough ranking quality to support downstream targeting. Aim for ROC-AUC >= 0.82, PR-AUC >= 0.45, and precision in the top 10% scored sessions >= 0.55.

Constraints

• Predictions must be generated daily in batch for marketing and lifecycle workflows.
• The model should be explainable enough for analysts to interpret key drivers.
• Avoid leakage from post-session events.
• Retraining should be feasible on a monthly cadence using standard Python tooling.

Deliverables

1. Define the target, leakage risks, and train/validation/test split strategy.
2. Build and compare at least two classification models.
3. Describe feature engineering and missing-data handling.
4. Evaluate the model using business-relevant ranking and classification metrics.
5. Recommend a production-ready model and explain tradeoffs.
6. Show Python code for preprocessing, training, tuning, and evaluation.