This repository contains the code for a final project in UC Berkeley, Fall 2021, IEOR 242 taught by Paul Grigas. In this project, we model the number of bicycle rentals per hour in the Bay Wheels service for 2018 and the first half of 2019 with the goal of building robust forecasting models. We also explore the spatiotemporal approach of forecasting bike rental density over time.
We present code and results for Linear Regression, CART, Random Forest, Gradient Boosting, and Recurrent Neural Network models for total Bay Wheels bicycle rentals per hour. We further provide a more general spatiotemporal analysis with an Random Forest model that aims to predict rentals at each rental location per day.
Contributors: Newton Cheng, Juliette Chevri`ere, Yiman Hu, Joshua Jacob, & Xingchen Liao
Data can be downloaded from
Run Exploratory Data Analysis.ipynb
Run data_preprocessing.ipynb
to add timestamp and zipcode for raw data collected.Run data_aggregation.ipynb
to aggregate all data to a single dataset.Run time_series_features.ipynb
to aggregate time series data with the overall dataset.
Run statistical_models.ipynb
and RNN Model.ipynb
to reproduce the models.
Run .ipynb
s in Spatiotemporal Analysis directory