About the Dataset

The dataset has a combination of quantitative and qualitative variables including:

Weather Conditions:

Quantitative: Temperature, Windspeed, Humidity
Qualitative: 4 Types of Weather conditions (Sunny, misty, light rain/snow, heavy rain/snow)

Time of the ride:

Quantitative: Year, Month, Day, Hour
Qualitative: Workday, Holiday

Ride numbers:

Registered/Casual

Research Questions

What is the general trend of bike-shares overtime? Does it generally increase, decrease, or stay the sam
What are some patterns of the bike-share rides? Does it varies between day types, hours, weather, season, and year?

Methodology

Preliminary Analysis: Weather Conditions & Day Types

Categorical Variables

Start of the game: human body

Day Types:

Significant difference btw holiday/non-holiday
Different but not so much btw workday/non-workday

Weather Types:

Each weather condition is significantly different from each others

Quantitative Weather Variables: Humidity, Tempeterature, Temperature Feel

Start of the game: human body

Negative trend between Humidity & Rides
Positive trend between Temperature & Rides
Positive trend between Windspeed & Rides, though a flatter slope

Dataset Cleaning

Correlation Graph

Temp and TempFeel (feel like temperature) are highly correlated, so I took away TempFeel in the final dataset.

Before:
Raw ride information

After:
Date information is summarized into categorical variables: day type, season, and time of day. Total number of rides in a day are normalized through square root transformation.

Bike Share Trends

Overall Trend

Smoothed Trend of Daily Rides

Day Types:

An overall increase from 2011 to 2012
Peak in summer, down in winter

Quantitative Weather Variables: Humidity, Tempeterature, Temperature Feel

Average Total Rides by Hour, Compared across Daytypes

Negative trend between Humidity & Rides
Positive trend between Temperature & Rides
Positive trend between Windspeed & Rides, though a flatter slope

Ride Prediction through Linear Regression Analysis

Adjusted R-square of the model is 0.66669, showing that overall, the model is a good fit. Below is an overview of the model output: most variables are highly significant to total ride (sqrt) prediction. Temperature and Time of Day, who has the largest coefficients, are the strongest variables.

Year

Significant rideshare growth from 2011 to 2012

Weather

Relative to Clear, Sunny weather, small drop for mist/cloudy weather
Big drop for light rain/light snow
Significant negative impact for heavy rain/snow/ice pellet days

Day Type

Weekend rides are not significantly different from weekdays
Holidays rides are less than weekdays & weekends

Season

Fall has most rides, followed by spring, summer, winter has the least rides

Time of Day

Compared to midnight (00-03), afternoon (16-19) has the largest rides (commute peaks), followed by mornings (08-11) (morning commute)

Variable Importance Analysis using Decision Tree Model

Adjusted R-square of the model is 0.66669, showing that overall, the model is a good fit. Below is an overview of the model output: most variables are highly significant to total ride (sqrt) prediction. Temperature and Time of Day, who has the largest coefficients, are the strongest variables.

Level 1 - Time of Day

Time of day is the primary driver:Most rides are between 08-23

Level 2 - Temperature

For daytime/evening, temperature is next
On cool days, late evening (8–11pm) lifts demand
Warmer days has significantly more rides than cooler days

Level 3 - Year

Lastly, the year is the deciding variable

Conclusion