BiciMAD Bike-Sharing System Dataset Creation and Preliminary Hourly Demand Prediction using Machine and Deep Learning Approaches
Group:
- Erika Gutierrez
- Javier Roset Cardona
In this project we create station level hourly demand forecasting models for BiciMAD, Madrid's bikeshare system. We first aggregated data from multiple open source online data portals into a cohesive dataset. The dataset contains hourly level entries from January 1st, 2019 to December 31st, 2022 for all stations in the bikeshare system resulting in 8.1 million rows. We then test the performance of an XGBOOST and a Graph Convolutional Neural Network on this dataset to generate bike demand estimates.
A summary of our project can be found here.
We have three main folders: data, models, and processing
In each of these folders there is a development folder that contains the notebooks we used to explore the data and test code that would be used in our actual ML pipeline.
data
- The folder
downloadingcontains the code we use to download data from their respective online sources in their raw form. - The folder
processingcontains the code we use to take this raw data, clean it, and combine it.
processing
- The folder
Feature Engineeringcontains the code we used to create the final dataframe of our features and targets. Also contains the script to collect the data used to populate our adjancency matrix.
models
- contains the various scripts used for creating, training, testing, and assessing our models. The
baseline_modeling.pycontains the code for the baseline and XGBOOST model andGNN_Main.pycontains the code for the GCNN model.