Detect Anomalous Trip Behavior

# Detect Anomalous Trip Behavior Develop a process to identify days, times, or locations where green-taxi traffic or behavior deviates from its normal pattern. You may use time-series methods, clustering, or another justified approach. Explain how you validate anomalies and investigate possible causes without treating correlation as proof. ### Constraints & Assumptions - Normal traffic has strong hour-of-day and day-of-week seasonality. - Data outages, schema changes, and invalid records can look like real-world anomalies. - The process should rank findings for review rather than guarantee a causal explanation. ### Clarifying Questions to Ask - Which metric matters: trip count, fare, distance, speed, location mix, or several? - What spatial and temporal resolution is actionable? - Are known holidays, weather, or service incidents available for validation? ```hint Model expected behavior before residuals A busy hour is not anomalous if that hour is normally busy; compare it with an appropriate seasonal baseline. ``` ### What a Strong Answer Covers - Metric, grain, data-quality checks, and seasonal baseline. - Anomaly score and threshold tied to review capacity or impact. - Backtesting, false-positive analysis, and robustness across methods. - A disciplined path from detection to external evidence and explanation. ### Follow-up Questions - How would you detect a gradual regime shift rather than a one-time spike? - How would you prevent one data outage from triggering thousands of alerts?