Prophet was developed internally at Facebook to meet the demand for high level forecasting and still provide flexibility. Out of the box Prophet typically produces high level forecast which is great but it's the insights into the data while doing that which are gold. For instance
- User specified seasonaility curves such as hourly, quarterly, weekly, monthly or annually.
- Better understand effect of holidays on demand. Qunatified in percentage change.
- Positive or negative trend once seasonality is removed from data
We will be using data from Bike Share Program in Chicago, Illinois. The data contains the number of bicycle Rides taken each our from the beginning of 2014 through the end of 2017.
Our objective is to forecast for 2018, and while we are at It we are going to analyze our data for demand trends during weekly, monthly and holidays.
Forecast for 2018
Overall Trend is Positive. We see a steady increase in rideshare since 2014
Annual Seasonality
- We see demand for bikes begins to increase in May (Spring) and starts to decline in November. Demand conincides with Summer and declines in Winter.
Weekly Seasonality
- Ridership peaks on Tuesday with 20% increase and then gradually declines over the week and disappears on weekends. It shows that there is different seasonlity for week and weekends
Demand during holidays
Lets convert the above output into numbers which help us see how ridership demand looks like in each of the holiday.
We see that demand is negative on all holidays except three
- Casmir Pulaski Day
- Columbus day
- Election Day