Predict Tip Percentage

# Predict Tip Percentage Create a derived target for tip as a percentage of total fare, then design and evaluate a predictive model for that target using green-taxi trip records. Explain feature timing, data cleaning, validation, performance metrics, and limitations. ### Constraints & Assumptions - Define behavior when total fare is zero, negative, missing, or internally inconsistent. - Do not use information unavailable at the intended prediction time. - Tip recording may depend on payment type, so missing or zero tips may not mean the same thing for every trip. ### Clarifying Questions to Ask - Is prediction made before pickup, during the trip, or after fare calculation? - Does total fare include the tip, and should the denominator instead be pre-tip charges? - Is the goal point prediction, ranking, or identifying likely high-tip trips? ```hint Audit the target before modeling The denominator and payment mechanism can create extreme values, censoring, or leakage before any model is trained. ``` ### What a Strong Answer Covers - A precise target and invalid-denominator policy. - Leakage-safe features and temporal or grouped validation. - Simple baselines, suitable losses and metrics, and calibration or residual analysis. - Segment performance, interpretability, and responsible use. ### Follow-up Questions - How would you model the large mass at zero? - What drift would you expect if payment behavior changes?