bixi-from-gas-price

• 

  The data sets

• 

  The model is fairly good...

• 

  ...but increasing the gas price by $1/L produces strange answers (negative ridership, etc)

Bixi use and gas prices: negative result

We tried to model how much Bixi use would change when gas prices changed, but we could not detect very much change based on the data we had. See the pictures.

Tool used

XGBoost.

Note that the repository uses Git LFS (Large File Storage), so you would probably need to install that to clone it.

Data sets

We spent a lot of time trying to get data sets, and eventually got:

• Gas price data set
• Bixi dataset
• Weather
• Cars waiting at intersections

The last data set turned out not to be useful for this purpose, because the intersection changed arbitrarily every day, so a fine-grained time-series comparison was not really possible. However, we used the first three to look for dependence of Bixi use on gas price, controlled by weather and time.

Explanation

We tried to analyse 2016-2017, which was available in both the Bixi dataset and the gas price data set, but unfortunately there were no large gas price changes in that time. The model was able to predict the Bixi use fairly well from time of day, weekday, temperature, and weather description (sunny, cloudy...), but it did not see much dependence on gas price.

Built With

• jupyter
• xgboost