Forecasting Demand for Ridshare Companies

-- Project Status: [Completed]

Project Objective

This project aims to help rideshare providers forecast demand, aiming to increase ridership and reliability for customers. The data was retrieved from the Chicago Data Portal

Methods

• Querying Data:
  • Pulling data from Socrata Open Data Portal
• Cleaning Data:
  • grouping by timeframes, computing null values
• Time Series Modeling:
  • gathering trends of data to forecast
• Plotting:
  • visualizing data and forecast results
• Multi-Core Processing:
  • utilizing all cores to improve run-time of calculations

Including Packages

• requests - query SODA database
• pandas - loading and pre-processing data
• matplotlib, seaborn - plotting results
• statsmodels, pmdarima - time series modeling
• kepler.gl - geospatial visualizations

Project Description

This project is broken down into four jupyter notebooks:

• query_clean:

  To start, using the SOCRATA platform, I grouped the data by ridership on a hourly basis, as the original file by ride is too large to hold on a local machine. I gathered information about my dataset, including days and hours of peak ridership, community areas with the highest ridership, and weather on a daily basis. These numbers improve understanding of the dataset, and help decision making for further modeling.

• city_modeling:

  This consists of running models on a 'city-wide' basis, to determine base scores and best model for the data. Tests include SARIMA, auto-SARIMA, and Holt-Winters' Exponential Smoothing. After optimizing parameters, I chose to continue with Holt-Winters' because of the superior speed, accuracy, and minimal computation needed.

• neighborhood_models:

  To develop accurate location informed decisions, I separated the data into individual community areas, and develop separate models, as demand differentiates for each. Modeling for each neighborhood is computationaly expensive, so incorporating multi-core processing brought modelin time from 1 hour to ~10 minutes.

• kepler.gl

  The ultimate output of this project would be to visualize when and where demand was happening. Kepler as a tool was able to combine geospatial boundaries for community areas along with the time-series of ridership to accurately represent demand - viewable on this Live Dashboard

Contact

• Feel free to contact me if any questions, or comments on the project - [hello@dancorley.com]