🔬 Bike Sharing Diagnostics

Overview

Public dataset from Capital Bikeshare, Washington D.C. (2011–2012): 731 daily and 17,379 hourly records of registered vs. casual riders.

Data Structure

Level	Records	Features
Daily	731	Total/registered/casual riders, weather, temperature, humidity, holiday
Hourly	17,379	Same features at hourly granularity: rush-hour patterns visible

Causal Reasoning Model

The diagnostic engine performs cross-level drill-down: starting from a daily anomaly, it zooms into hourly rush-hour patterns to pinpoint when during the day demand was unusually low or high.

Key factors in the causal model:

• Weather: rain, snow, fog, extreme heat/cold
• Temperature: normalized feel temperature vs. actual
• Humidity: high humidity suppresses ridership
• Holidays: reduced commuter demand
• Seasonality: summer peaks, winter troughs

Sample Questions

• "Why was bike demand low on 2012-10-29?" (Hurricane Sandy)
• "What hourly pattern explains the daily anomaly?"
• "How did weather suppress ridership this week?"