Turnstile - IEMS 303 Final Presentation

Predicting CTA Ridership with multivariate regression for Statistics I at Northwestern University

Transcript

1. Predicting CTA Ridership with multivariate regression Team 1: Brian Lange Jenny Park Justice Pettigrew Sean McGinnis
2. Project Overview Model and predict future ‘L’ weekday ridership based on existing ridership data from previous years Data Collection Data Analysis Predict Ridership on the CTA
3. Other factors to consider: Consistent outliers (gay pride parade, etc.) Weekends and Holidays Local causes of variation (schools, stadiums, etc.) Null Hypothesis: Average weekday ridership does not display seasonal patterns or changes in ridership over time Hypothesis Testing Alternative Hypothesis: Average weekday ridership does display seasonal patterns or changes in ridership over time
4. Focusing on Commuters Data Collection & Processing Removed weekends and holidays Removed outlier weekdays (>2 normalized residuals away from predicted value) Pre-gathered Data https://data.cityofchicago.org/ Number of riders going through the turnstiles at each stop, for each day since 2001 to 2013
5. Python Programming Python code: Calculates regressions • (coefficients, p-values, r2 values, scatter plots, etc.) Too many regressions: 144 stops x 2 = 288 regression calculations using 652,291 rows of data statsmodels
6. Regression Model Example - Monroe Riders = 1005.01 + -0.62828 • day + 0.00019771 • day2 + -66.411 • month + 22.905 • month2 - 1.4729 • month3 R-sq (adj) = 62.62%
7. Benefits to Society CTA: Meet Ridership Demand Commuters: Predict Ridership
8. Next Steps Account for further factors Explore why outlier stops do not fit the model
9. Questions? Source code and plots at github.com/bjlange/turnstile
10. Appendix Slides: Rides = 4514.4 - 0.424293 day +0.00014457 day^2 R-Sq(adj) = 3.16% Initial Regression Model
11. Appendix Slides Outlier Stops
12. Limitations: Causes of Variation Weekly fluctuations Local Factors (Schools, Stadiums, etc.) Trending neighborhoods Future projects including new buildings or changes to CTA policies can not be predicted