Business Context

EcoMat Labs is developing a new biodegradable polymer for food packaging and wants to predict its 12-month breakdown rate before running long, expensive field studies. You need to build a regression model that estimates the percentage of mass lost after 12 months under different material formulations and environmental conditions.

Dataset

You are given historical lab and field trial data from prior biodegradable polymer experiments, including formulation chemistry, processing settings, and exposure conditions.

• Size: 18,500 polymer trial records, 44 features
• Target: Continuous — percentage mass loss after 12 months
• Missing data: ~8% missing in environmental sensor readings, ~3% missing in lab-measured material properties
• Target distribution: Continuous, moderately right-skewed, with most samples between 15% and 70% mass loss

Success Criteria

A good solution should achieve MAE below 6 percentage points on a held-out test set and provide enough interpretability for materials scientists to understand which formulation and environmental variables drive degradation.

Constraints

• Predictions will be used in R&D screening, so interpretability matters
• Inference is batch-only and can run offline weekly
• The model should generalize to new polymer formulations, not just memorize batch history
• Avoid leakage from repeated measurements of the same formulation family across splits

Deliverables

1. Build a regression model to predict 12-month breakdown rate
2. Define what data is required and justify why each feature group matters
3. Explain preprocessing, feature engineering, and split strategy
4. Compare at least one baseline model with a stronger nonlinear model
5. Report evaluation metrics and discuss production tradeoffs