Seattle Fremont bridge bicycle daily count forecast using facebook's prophet library [project]

Project Aim: The aim is to forecast the Fremont Bridge Bicycle ridership for next two years using facebook's prophet library.

Requirements: python 3.9, numpy, matplotlib, plotly, prophet and json

OS: Tested on windows and Linux

Dataset Description

The Fremont Bridge Bicycle Count dataset includes bycycle count from October 2012 to 2019. It consist of count records (number of bikes that cross the bridge using the pedestrian/bicycle pathways). Inductive loops on the east and west pathways count the passing of bicycles regardless of travel direction.

Notebook Information

• This notebook consist of forecasting codes
• Forecasting is done using Meta's prophet library.
• Here I have forecasted the Bicycle counts for next 2 years.
• Included holidays (Washinton) to observe the impact on daily counts.
• Performed a forward-chaining cross-validation.

Project outline step by step:

1. Importing important libraries
2. Reading and loading data from kaggle: Data Source
3. EDA: Performing exploratory analysis
4. Converted (resamples) data to daily and weekly count to check the pattern and inconsistencies

Here is a plot for daily ridership count:

5. Next, added holidays to check the impact of holidays on count forecast
   • Holidays associated with Washinton states has been utilised
6. Trained the model using prophet library
   • Set the seassionality mode to multiplicative
   • Set a yearly seassionality to 4
   • Added the holidays
7. Generated the forecast plot using matplotlib and plotly (interactive plot)

8. Saved the model to json file using pickle library
9. Generated a components plot using matplotlib and plotly (interactive plot)
   • The component plot showed that the bicycle ridership tread is upward since 2017 and still increasing.
   • The yearly trend showed that June to August (summer) the bicycle ridership grows by 40%, while it drops by 20%-40% during January to March (winder season).
   • Weekly ridership trend revealed that Monday to Friday the ridership increases by 20% while during weekend it drops by 40%.

10. Performed model evaluation using a forward chaining cross-validation with
    • horizon = '90 days'
    • period = '30 days',
    • initial = '730 days'
    • parallel = 'processes' (enabled parellel processing)
11. Generated a cross-validation metric plot
    • Used rmse as evaluation metric and selected a rolling_window of 0.1 (10%)