The data consists of more than 2.5 million trips taken by shared bikes, including start hour, start weekday, start month, station ids, customer types and other qualities related to a bike sharing system in a large city. The dataset for the San Francisco Bay area can be found here, with feature documentation available here. The cleaned dataframe can be downloaded here.
As I wanted to cover a full year with analysis, I downloaded 17 .csv files which were merged into one large .csv file. After having assessed the dataset visually and programmatically, I decided that I needed to perform some data wrangling in order to use it for my analysis. As a result, I converted the features birth years, and station ids from float to int. However, I had to drop the rows where these columns contained null values. Additionally I wanted to investigate the time feature, so I converted the start and end time columns to datetime. Then I created addititional columns
start_month,start_weekday, andstart_hourwhich stored corresponding values that were extracted from thestart_timefeature. Two further columnsstart_weekday_nameandstart_month_namewere created that stored not the corresponding numbers of the weekdays and months but their names. What's more, I created the columndifference_stationwhich stored the result of the calculationstart_station_id-end_station_id.
In the exploration, I found that there was a strong relationship between the duration in seconds and the difference in station ids, start station ids and the year of birth of the members. A smaller difference in station ids was usually associated with a longer duration in seconds. The same was true when looking at start station ids. Additionally, the year of birth of the customers showed a strong logarithmic effect on the duration in seconds. In general, younger customers tended to rent a bike for a longer period of time.
When investigating the start time together with the customer type, I found out that (irregular) customers rented a bike for a much longer period of time on average compared to subscribers. What's more, the relationship between the start weekday and the station id showed that on weekends the distribution between the station ids is more balanced compared to workdays where mostly the smaller station ids are well frequented.
For the presentation, I focus on what influence the start time, the customers as well as the station ids will have on the duration of a trip and leave out the other variables. I start by introducing the duration variable, followed by the start time variable which will be depicted as a distribution of hour, then plot the transformed heatplot.
Afterwards, I introduce the station id variable. To start, I'm plotting a two dimensional histogram which shows the effect of the difference in stations on the duration.
Another main categorical variable will be the weekday which will be presented by a distribution plot followed by showing its influence on the stations.
The last main categorical variable will be the user type which will be presented by showing its influence on the duration, start hour, start station id, and the weekdays. These plots will depict multivariate distributions.