MikeYuanMY / Capital_Bikeshare_Analysis

Capital Bikeshare Ridership Prediction

^{Mike Yuan (Mikeymice) and Evan Yathon (EvanYathon)}

Data Citation:

[1] Fanaee-T, Hadi, and Gama, Joao, "Event labeling combining ensemble detectors and background knowledge", Progress in Artificial Intelligence (2013): pp. 1-15, Springer Berlin Heidelberg, doi:10.1007/s13748-013-0040-3.

Bike sharing systems are becoming one of popular transportation measures in the urban areas. The bike share systems usually provide customers with durable bikes and stations for rental and parking. The bikes can be unlocked from any station and returned to any station at any time.

We are interested in answering the question.

What are the top three predictors of rental bike ridership?

To answer this question, we used data originally sourced from Capital Bikeshare in metro DC. We performed supervised machine learning using decision tree classification. Valid features such as temperature, weather and day of the week were included. In order to simplify the potential ridership outcomes, the ridership was broken into different chunks; Low, Medium and High ridership categories. To find the bounds of the categories, we used quantiles to evenly split the number of rentals per day in our data. Data cleaning and wrangling were performed.

To optimize the accuracy of the decision tree, we performed cross validation on the cleaned data to define the ideal maximum height for the tree. The decision tree was trained using this optimal height. By visualizing the tree and utilizing scikit-learn's feature_importances function, we determined which of the features are the strong predictors of daily bicycle ridership.

Link to report.

Data Attributes

This analysis was performed on the Bike Sharing Dataset provided by the UCI Machine Learning Repository

Both hour.csv and day.csv had the following fields, except hr which was not available in day.csv

Variable	Description
instant	record index
dteday	date
season	season (1:springer, 2:summer, 3:fall, 4:winter)
yr	year (0: 2011, 1:2012)
mnth	month ( 1 to 12)
hr	hour (0 to 23)
holiday	weather day is holiday or not
weekday	day of the week
workingday	if day is neither weekend nor holiday is 1, otherwise is 0.
weathersit	Clear, Few clouds, Partly cloudy, Partly cloudy Mist + Cloudy, Mist + Broken clouds, Mist + Few clouds, Mist Light Snow, Light Rain + Thunderstorm + Scattered clouds, Light Rain + Scattered clouds Heavy Rain + Ice Pallets + Thunderstorm + Mist, Snow + Fog
temp	Normalized temperature in Celsius. The values are derived via (t-t_min)/(t_max-t_min), t_min=-8, t_max=+39 (only in hourly scale)
atemp	Normalized feeling temperature in Celsius. The values are derived via (t-t_min)/(t_max-t_min), t_min=-16, t_max=+50 (only in hourly scale)
hum	Normalized humidity. The values are divided to 100 (max)
windspeed	Normalized wind speed. The values are divided to 67 (max)
casual	count of casual users
registered	count of registered users
cnt	count of total rental bikes including both casual and registered

Usage

1. Clone this repo, and using the command line navigate to the root of this project.

2. Run the following command to produce the report:

   make all

   or on your command shell

   bash run_all.sh

   Run the following command to clean previous result

   make clean

3. The report is generated under the report/ directory

Docker

To run this analysis using Docker, clone/download this repository, use the command line to navigate to the root of this project on your computer, and then type the following (filling in PATH_ON_YOUR_COMPUTER with the absolute path to the root of this project on your computer).

To pull the docker image

docker pull yuanmike/capital_bikeshare_analysis

To generate the report

docker run --rm -v PATH_ON_YOUR_COMPUTER:/home/capital_bikeshare_analysis yuanmike/capital_bikeshare_analysis make -C 'home/capital_bikeshare_analysis' all

To clean the results

docker run --rm -v PATH_ON_YOUR_COMPUTER:/home/capital_bikeshare_analysis yuanmike/capital_bikeshare_analysis make -C 'home/capital_bikeshare_analysis' clean

Workflow

Our workflow was the following:

1. Perform data cleaning and feature engineering
2. Perform Machine Learning with Decision Tree and output feature importances
3. Use ggplot2 to produce EDA graphs
4. Construct report with results from EDA and the Decision Tree model

In order to reproduce our findings, both Makefile and run_all.sh took the same work flow to generate the report. The scripts would run load_csv.R to generate cleaned data. summarise_data.py would perform data analysis and produced result data. eda_plot.R would use cleaned data to generate plots and importance_features.R would also plot the feature importances from the result. Finally, with knitr, capital_bikeshare_analysis.Rmd would generate our final report in the markdown format.

Here is a dependency graph of our outputs, scripts and inputs used in Makefile.

V2.0 Dependencies

• R and R libraries (R Version 3.5.1)

  • tidyverse v1.2.1
  • lubridate v1.7.4
  • ggplot2 v3.0.0
  • rmarkdown v1.10
  • knitr v1.20
  • pracma v2.1.8

• Python (Python Version 3.6.5)

  • pandas v0.23.0
  • numpy v1.14.3
  • sklearn v0.19.1
  • graphviz v0.10.1
  • pydotplus v2.0.2
  • seaborn v0.9.0
  • Ipython v6.4.0
  • matplotlib v2.2.2
  • tqdm v4.28.1