Business Context

CivicStat, a public sector analytics team supporting a mid-sized city, wants to forecast monthly service demand at the neighborhood level so agencies can allocate inspectors, sanitation crews, and outreach staff more effectively. You need to build a regression model that predicts next month's non-emergency service request volume.

Dataset

You are given 48 months of neighborhood-level data from 220 neighborhoods. Each row represents one neighborhood-month.

• Rows: 10,560 neighborhood-month observations, 33 features
• Target: Continuous — next month's service request count per neighborhood
• Missing data: ~12% missing in operational staffing fields, ~4% missing in demographic refresh fields
• Skew: Request counts are right-skewed, with a small number of high-demand neighborhoods

Success Criteria

A strong solution should achieve MAE below 18 requests, RMSE below 30, and outperform a seasonal baseline that predicts the same month last year's volume.

Constraints

• Predictions are generated monthly in batch; inference should finish in under 5 minutes for all neighborhoods.
• City stakeholders need interpretable drivers of demand, not just predictions.
• The model should be robust to policy changes and seasonal effects.

Deliverables

1. Build a regression pipeline to predict next-month service request volume.
2. Explain model choice and how you handle skew, missing values, and mixed feature types.
3. Compare against at least one simple baseline.
4. Report evaluation metrics on a time-based holdout set.
5. Provide feature importance or coefficient-based interpretation for agency stakeholders.