Bike Sharing: A study investigating the impact of external conditions on bike rental counts

Abstract:

​

This paper explores the factors that significantly influence the number of bike rentals made per day by registered and casual/unregistered users, using the data collected from the Capital Bikeshare System spanning across the years 2011-2012. Furthermore, a key objective was to gain insight into the usage patterns of these groups and to what extent factors affect them differently. Following a preliminary investigation of variable correlations and trends, six models were selected with different combinations of variables and quadratic transformations. A quadratic and linear model were trained on the 2011 data (2012 served as the validation set), and while all were significant, diagnostic analysis of their residuals and normality presented concerns of heteroskedasticity as well as some deviation from normality. This violated key model assumptions. Performance on the validation set was measured by MSE and Relative MSE, and analyzed in greater detail through predictive plots. Error was greater than expected, specifically for the registered model; while relative trends seemed to match up, the model significantly underfit. This result is likely due to growing demand for bike sharing rentals, moreover an increase in the company’s customer base, which cannot be accounted for by weather or workday factors. Nonetheless our best performing model was fairly accurate in predicting the relative trends of these two user groups, particularly those of casual riders who appeared to be influenced to a larger extent by the considered factors, but multi-year data would substantially approve the ability to account for company growth.

​

​

Here is a link to the paper. This was a collaborative effort which I undertook with teammates: Fahar Laqa, Ryan Shnitman, Shuman Jiang and Keifei Wang.

​

Here is a presentation given by the entire team summarising our efforts throughout the semester.