Case Study (R): bike-sharing service data analysis

mean(ds_all_trips_clean_length_correct$ride_length_hours) #hours straight average (total ride length hours / rides)
mean(ds_all_trips_clean_length_correct$ride_length_mins) #mins straight average (total ride length minutes / rides)
mean(ds_all_trips_clean_length_correct$ride_length_secs) #secs straight average (total ride length seconds / rides)

median(ds_all_trips_clean_length_correct$ride_length_hours) #midpoint number in the ascending array of hour ride lengths
median(ds_all_trips_clean_length_correct$ride_length_mins) #midpoint number in the ascending array of mins ride lengths
median(ds_all_trips_clean_length_correct$ride_length_secs) #midpoint number in the ascending array of secs ride lengths

max(ds_all_trips_clean_length_correct$ride_length_hours) #hours longest ride
max(ds_all_trips_clean_length_correct$ride_length_mins) #mins longest ride
max(ds_all_trips_clean_length_correct$ride_length_secs) #secs longest ride

min(ds_all_trips_clean_length_correct$ride_length_hours) #hours shortest ride
min(ds_all_trips_clean_length_correct$ride_length_mins) #mins shortest ride
min(ds_all_trips_clean_length_correct$ride_length_secs) #secs shortest ride

summary(ds_all_trips_clean_length_correct$ride_length_hours)
summary(ds_all_trips_clean_length_correct$ride_length_mins)
summary(ds_all_trips_clean_length_correct$ride_length_secs)

N.B. Between mean and median, it is preferable to consider the median.
The average is usually the most appropriate measure for determining a position. This is because it takes into account every value in the dataset. However, outliers in the dataset may affect the mean leading it not to accurately represent all scores (as is the case in the dataset used in this case study). Thus, the median is better because outliers do not affect it.

setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_hours ~ ds_all_trips_clean_length_correct$member_casual, FUN = mean), c("customers", "mean hours"))
setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_mins ~ ds_all_trips_clean_length_correct$member_casual, FUN = mean), c("customers", "mean mins"))
setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_secs ~ ds_all_trips_clean_length_correct$member_casual, FUN = mean), c("customers", "mean secs"))

setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_hours ~ ds_all_trips_clean_length_correct$member_casual, FUN = median), c("customers", "median hours"))
setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_mins ~ ds_all_trips_clean_length_correct$member_casual, FUN = median), c("customers", "median mins"))
setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_secs ~ ds_all_trips_clean_length_correct$member_casual, FUN = median), c("customers", "median secs"))

setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_hours ~ ds_all_trips_clean_length_correct$member_casual, FUN = max), c("customers", "max hours"))
setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_mins ~ ds_all_trips_clean_length_correct$member_casual, FUN = max), c("customers", "max mins"))
setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_secs ~ ds_all_trips_clean_length_correct$member_casual, FUN = max), c("customers", "max secs"))

setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_hours ~ ds_all_trips_clean_length_correct$member_casual, FUN = min), c("customers", "min hours"))
setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_mins ~ ds_all_trips_clean_length_correct$member_casual, FUN = min), c("customers", "min mins"))
setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_secs ~ ds_all_trips_clean_length_correct$member_casual, FUN = min), c("customers", "min secs"))

setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_secs ~ ds_all_trips_clean_length_correct$member_casual + ds_all_trips_clean_length_correct$day_of_week, FUN = mean), c("customers", "day week", "duration in seconds"))

4.1 – Note that the days of the week are not in order; let’s fix that.
Indicate the names of the days in Italian if the operating system is in this language (provided that you are using an application installed on your computer for processing this work, e.g., RStudio Desktop); otherwise, use the relevant translation (on online platforms, leave the names of the days in English):

ds_all_trips_clean_length_correct$day_of_week 

4.2 – Let us analyze the average travel time for each day among subscribers and occasional users again:

setNames(aggregate(ds_all_trips_clean_length_correct$ride_length_secs ~ ds_all_trips_clean_length_correct$member_casual + ds_all_trips_clean_length_correct$day_of_week, FUN = mean), c("customers", "day week", "duration in seconds"))

To avoid the `summarise()` message grouped output by `member_casual`, we can hide it by using the `.summarise` argument and setting it to “FALSE“.

options(dplyr.summarise.inform = FALSE)

CAUTION: On all graphs, the “lubridate” function was used to sort the data with the day of the week in the language according to the local time zone, which in my case, starts on Monday (otherwise, the list would start on Sunday). If it is necessary to respect the original ordering remove lubridate (basically, remove the single instruction with lubridate and reactivate the next one disabled with the # symbol).

ds_all_trips_clean_length_correct %>% 
 mutate(weekday = lubridate::wday(ds_all_trips_clean_length_correct$started_at, label = TRUE, week_start = getOption("lubridate.week.start", 1),locale = Sys.getlocale("LC_TIME"))) %>% #creates weekday field using wday()
 #mutate(weekday = wday(started_at, label = TRUE)) %>%
 group_by(member_casual, weekday) %>% #groups by usertype and weekday
 dplyr::summarise(number_of_rides = n() #calculates the number of rides and average duration
 ,average_duration_hours = mean(ride_length_hours), average_duration_mins = mean(ride_length_mins), average_duration_secs = mean(ride_length_secs)) %>% #calculates the average duration
 arrange(member_casual, weekday) #sorts

# This function help to resize the plots
function_help_resize_plots 

ds_all_trips_clean_length_correct %>% 
 mutate(weekday = lubridate::wday(ds_all_trips_clean_length_correct$started_at, label = TRUE, week_start = getOption("lubridate.week.start", 1),locale = Sys.getlocale("LC_TIME"))) %>%
 #mutate(weekday = wday(started_at, label = TRUE)) %>% 
 group_by(member_casual, weekday) %>% 
 dplyr::summarise(number_of_rides = n(), average_duration = mean(ride_length_secs)) %>% 
 arrange(member_casual, weekday) %>% 
 ggplot(aes(x = weekday, y = number_of_rides, fill = member_casual)) +
 guides(fill = guide_legend(title = "User category")) +
 scale_y_continuous("number of runs", labels = scales::comma) +
 scale_x_discrete(name = "day week") +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 geom_col(position = "dodge")

Some considerations can be drawn from the graphs just compiled:

• during the weekend (Saturday and Sunday), occasional users significantly increase the number of rides compared to other single days of the week.
• annual subscribers make more trips during the work week than during the weekend, so behavior related to use of the subscription for work purposes (home/work commute) is presumable.

ds_all_trips_clean_length_correct %>% 
 mutate(weekday = lubridate::wday(ds_all_trips_clean_length_correct$started_at, label = TRUE, week_start = getOption("lubridate.week.start", 1),locale = Sys.getlocale("LC_TIME"))) %>%
 #mutate(weekday = wday(started_at, label = TRUE)) %>% 
 group_by(member_casual, weekday) %>% 
 dplyr::summarise(number_of_rides = n(), average_duration = mean(ride_length_hours)) %>% 
 arrange(member_casual, weekday) %>% 
 ggplot(aes(x = weekday, y = average_duration, fill = member_casual)) +
 guides(fill = guide_legend(title = "User category")) +
 scale_y_continuous("average duration in hours") +
 scale_x_discrete(name = "day week") +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 geom_col(position = "dodge")

ds_all_trips_clean_length_correct %>% 
 mutate(weekday = lubridate::wday(ds_all_trips_clean_length_correct$started_at, label = TRUE, week_start = getOption("lubridate.week.start", 1),locale = Sys.getlocale("LC_TIME"))) %>%
 #mutate(weekday = wday(started_at, label = TRUE)) %>% 
 group_by(member_casual, weekday) %>% 
 dplyr::summarise(number_of_rides = n(), average_duration = mean(ride_length_mins)) %>% 
 arrange(member_casual, weekday) %>% 
 ggplot(aes(x = weekday, y = average_duration, fill = member_casual)) +
 guides(fill = guide_legend(title = "User category")) +
 scale_y_continuous("average duration in minutes") +
 scale_x_discrete(name = "day week") +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 geom_col(position = "dodge")

ds_all_trips_clean_length_correct %>% 
 mutate(weekday = lubridate::wday(ds_all_trips_clean_length_correct$started_at, label = TRUE, week_start = getOption("lubridate.week.start", 1),locale = Sys.getlocale("LC_TIME"))) %>%
 #mutate(weekday = wday(started_at, label = TRUE)) %>% 
 group_by(member_casual, weekday) %>% 
 dplyr::summarise(number_of_rides = n(), average_duration = mean(ride_length_secs)) %>% 
 arrange(member_casual, weekday) %>% 
 ggplot(aes(x = weekday, y = average_duration, fill = member_casual)) +
 guides(fill = guide_legend(title = "User category")) +
 scale_y_continuous("average duration in seconds") +
 scale_x_discrete(name = "day week") +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 geom_col(position = "dodge")

Some considerations can be drawn from these graphs:

• occasional users have twice as much time using bicycles as annual subscribers;
• annual subscribers have their travel time equally distributed throughout the week;
• occasional ones see significantly higher usage (compared to other days of the week) on weekends and Mondays.

How much data concerns annual (member) subscribers, and how much concerns casual (casual) subscribers?

ds_all_trips_clean_length_correct %>% 
 group_by(member_casual) %>% 
 summarise(count = length(ride_id),
 '%' = (length(ride_id) / nrow(ds_all_trips_clean_length_correct)) * 100)

Data visualization with a bar graph:

function_help_resize_plots(16,8)
ggplot(ds_all_trips_clean_length_correct, aes(member_casual, fill=member_casual)) +
 guides(fill = guide_legend(title = "User category")) +
 geom_bar() +
 scale_y_continuous("number of runs", labels = scales::comma) +
 scale_x_discrete(name = "Occasionali vs Abbonati") +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 labs(title="Chart 1 - Casuals vs Members")

Visualization on a pie chart:

ds_all_trips_clean_length_correct %>% 
 group_by(member_casual) %>% 
 summarise(count = length(ride_id),
 percent_cat_users = round((length(ride_id) / nrow(ds_all_trips_clean_length_correct)) * 100, digits = 2)) %>% 
 #arrange(percent_cat_users) %>%
 #mutate(percent_cat_users_labels = scales::percent(percent_cat_users, accuracy = 0.01)) %>% 
 ggplot(aes(x = "", y = count, fill = member_casual)) +
 geom_col(color = "black") +
 geom_label(aes(label = paste0(percent_cat_users, "%")), position = position_stack(vjust = 0.5), show.legend = FALSE) +
 guides(fill = guide_legend(title = "User category")) +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 labs(title="Casuals vs Members (in %)",
 subtitle = "November 2021 - October 2022",
 x="",
 y="") + 
 #scale_fill_viridis_d() +
 coord_polar(theta = "y") +
 theme_void() # remove default theme

As seen in the “Occasional vs Subscribers” table, subscribers have a more significant proportion of the dataset, with a percentage of ~60%, ~20% more than the count of occasional cyclists.

How are the data distributed by month?

ds_all_trips_clean_length_correct %>%
 group_by(year_month) %>%
 summarise(count = length(ride_id),
 '%' = (length(ride_id) / nrow(ds_all_trips_clean_length_correct)) * 100,
 'members_p' = (sum(member_casual == "member") / length(ride_id)) * 100,
 'casual_p' = (sum(member_casual == "casual") / length(ride_id)) * 100,
 'dif_members_casuals' = members_p - casual_p)

ds_all_trips_clean_length_correct %>%
 ggplot(aes(year_month, fill=member_casual)) +
 geom_bar() +
 labs(x="Mese", title="Chart 2 - Distribution of trips per month") +
 guides(fill = guide_legend(title = "User category")) +
 scale_y_continuous("number of runs", labels = scales::comma) +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 theme(axis.text.x=element_text(angle=45,hjust=1))

Some considerations can be drawn from this graph:

• the number of runs seems to be affected by weather conditions;
• the month with the most significant number of data points was July 2022, with ~14.5% of the dataset;
• in all months, we have more subscriber runs than casual runs;
• the difference in proportion between annual and occasional subscribers is more minor in “hot” months;
• the distribution seems cyclical.

To test the correlation with climatic conditions, we compare the number of runs with the climatic data for Chicago during the case study period.

Source: Climate-Data.org (Daily average °C, November 2021 — October 2022).
Please note: If the monthly reference interval changes, it is necessary to manually change the temperature values in the following dataset “ds_chicago_temperatures“.

ds_stats_n_races <- ds_all_trips_clean_length_correct %>%
 group_by(year_month) %>%
 summarise(count = length(ride_id),
 '%' = (length(ride_id) / nrow(ds_all_trips_clean_length_correct)) * 100,
 'members_p' = (sum(member_casual == "member") / length(ride_id)) * 100,
 'casual_p' = (sum(member_casual == "casual") / length(ride_id)) * 100,
 'dif_members_casuals' = members_p - casual_p)
ds_chicago_temperature %
 ggplot(aes(year_month, fill=member_casual)) +
 geom_bar() +
 labs(x="Mese", title="Chart 2 - Distribution of trips per month") +
 guides(fill = guide_legend(title = "User category")) +
 scale_y_continuous("number of runs", labels = scales::comma) +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 theme(axis.text.x=element_text(angle=45,hjust=1))

ggplot(ds_chicago_temperature_races, aes(x = year_month, y = count)) + # bar plot
 geom_col(linewidth = 1, color = "darkblue", fill = "white") +
 scale_x_discrete(name = "year - month") +
 scale_y_continuous("number of runs", labels = scales::comma) +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 theme(axis.text.x=element_text(angle=45,hjust=1))

ggplot(ds_chicago_temperature_races, aes(x = month, y = temp_mean)) + # line plot
 geom_line(linewidth = 1.5, color="red", group = 1) +
 scale_x_discrete(name = "year - month") +
 scale_y_continuous(name = "medium temperature") +
 theme(axis.text.x=element_text(angle=45,hjust=1))

ggplot(ds_chicago_temperature_races, aes(x = year_month)) + 
 geom_col(aes(y = count), size = 1, color = "darkblue", fill = "white") +
 #scale_y_continuous("number of runs", labels = scales::comma) +
 geom_line(aes(y = 30000*temp_mean), linewidth = 1.5, color="red", group = 1) + 
 scale_x_discrete(name = "year - month") +
 scale_y_continuous(sec.axis = sec_axis(~./300000, name = "temp mean")) +
 theme(axis.text.x=element_text(angle=45,hjust=1),axis.text.y=element_blank(),axis.ticks.y=element_blank()) +
 labs(title = "In a comparison of stroke number and temperature", subtitle = "the red line indicates the average temperature in the corresponding month")

rm(ds_stats_n_races) # remove temporary dataset used to sync temperatures

The main result is:

• temperature affects the volume of monthly runs, particularly in winter/cold ones;
• when the temperature drops below freezing, occasional customers avoid bicycle use. Under the same conditions, annual subscribers maintain a specific frequency of use, probably related to an obligatory commute (home/work).

How are the data distributed over the days of the week?

ds_all_trips_clean_length_correct %>%
 group_by(weekday) %>% 
 summarise(count = length(ride_id),
 '%' = (length(ride_id) / nrow(ds_all_trips_clean_length_correct)) * 100,
 'members_p' = (sum(member_casual == "member") / length(ride_id)) * 100,
 'casuals_p' = (sum(member_casual == "casual") / length(ride_id)) * 100,
 'dif_perc_members_casuals' = members_p - casuals_p)

Let’s visualize the data via a graph (enlarge it for better reading)

ggplot(ds_all_trips_clean_length_correct, aes(weekday, fill=member_casual)) +
 geom_bar(position='dodge2') +
 #geom_text(stat = "count", aes(label = ..count..)) +
 geom_label(stat='count',
 aes(label=after_stat(count)),
 position=position_dodge2(width=0.5),
 size=3,
 show.legend = FALSE # removes the letter within the icon
  ) +
 labs(x="giorni della settimana", title="Chart 3.1 - Distribution by day of the week") +
 guides(fill = guide_legend(title = "User category")) +
 scale_y_continuous("number of runs", labels = scales::comma) +
 scale_fill_discrete(labels = c("Casuals", "Members"))
 #theme(legend.position = "none")

Same data but displayed differently:

ggplot(ds_all_trips_clean_length_correct, aes(weekday, fill=member_casual)) +
 geom_bar() +
 #geom_text(stat = "count", aes(label = ..count..)) +
 geom_text(stat = "count", 
 aes(label= after_stat(count)), 
 position = position_stack(vjust = 0.5),
 size = 3
  ) +
 labs(x="giorni della settimana", title="Chart 3.2 - Distribution by day of the week") +
 guides(fill = guide_legend(title = "User category")) +
 scale_y_continuous("number of runs", labels = scales::comma) +
 scale_fill_discrete(labels = c("Casuals", "Members"))

It is interesting to see:

• the volume of data is distributed essentially equally on all days of the week except Saturday;
• Saturday has the most data points;
• annual subscribers have the highest data volume, except on Saturdays. On this day of the week, casuals have the most data points;
• weekends starting on Friday see the highest volume of occasional rides, with an increase of up to 20 per cent.

ds_all_trips_clean_length_correct %>%
 group_by(start_hour) %>% 
 summarise(count = length(ride_id),
 '%' = (length(ride_id) / nrow(ds_all_trips_clean_length_correct)) * 100,
 'members_p' = (sum(member_casual == "member") / length(ride_id)) * 100,
 'casuals_p' = (sum(member_casual == "casual") / length(ride_id)) * 100,
 'dif_perc_members_casuals' = members_p - casuals_p)

ds_all_trips_clean_length_correct %>%
 ggplot(aes(start_hour, fill=member_casual)) +
 scale_y_continuous("number of runs", labels = scales::comma) +
 labs(x="Time of day", title="Chart 4.1 - Distribution by time of day") +
 guides(fill = guide_legend(title = "User category")) +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 geom_bar()

same results showed differently (enlarged graph)

ggplot(ds_all_trips_clean_length_correct, aes(start_hour, fill=member_casual)) +
 geom_bar(position='dodge2') +
 #geom_text(stat = "count", aes(label = ..count..)) +
 geom_label(stat='count',
 aes(label=after_stat(count)),
 position=position_dodge2(width=0.5),
 size=3,
 show.legend = FALSE # removes the letter within the icon
  ) +
 labs(x="time of day", title="Chart 4.2 - Distribution by time of day") +
 guides(fill = guide_legend(title = "User category")) +
 scale_y_continuous("number of runs", labels = scales::comma) +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 theme(legend.position = c(0.2, 0.8))
 # coord_flip()
 #theme(legend.position = "none")

From this graph, we can see:

• There is a more significant influx of cyclists in the afternoon;
• the number of rides of annual subscribers is significantly higher than those of occasional subscribers in the 6 to 10 a.m. and 3 to 7 p.m. time slots;
• casuals are more incredible than subscribers between midnight and 4 a.m.

This graph can be expanded by dividing it by each day of the week.

ds_all_trips_clean_length_correct %>%
 ggplot(aes(start_hour, fill=member_casual)) +
 geom_bar() +
 labs(x="Time of day", y="number of runs", title="Chart 5 - Distribution by time of day and day of the week") +
 guides(fill = guide_legend(title = "User category")) +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 facet_wrap(~ weekday)

There is a clear difference between weekdays and weekends.

We generate graphs for these two weekly splits.

ds_all_trips_clean_length_correct %>%
 mutate(type_of_weekday = ifelse(weekday == '6 - Sat' | weekday == '7 - Sun', 'weekend', 'midweek')) %>%
 ggplot(aes(start_hour, fill=member_casual)) +
 guides(fill = guide_legend(title = "User category")) +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 scale_y_continuous("number of runs", labels = scales::comma) +
 labs(x="Time of day", title="Chart 6 - Distribution by time of weekday-weekend") +
 geom_bar() +
 facet_wrap(~ type_of_weekday)

The two plots differ in some key aspects:

• while weekends have a steady flow of data points, midweek days have a steeper/irregular flow of data;
• counting data points is not very meaningful since each graph represents a different number of days;
• for midweek days there is a significant increase in data points between 6 and 8 a.m. (especially of subscribers) and then declines in the next two hours (9 and 10 a.m.) and then resumes with a steady increase from 11 a.m. to 6 p.m;
• another significant increase is from 4 p.m. to 6 p.m., particularly in subscribers;
• during the weekend, we have a greater flow of casual users between 11:00 am and 6:00 pm.

Summarizing the above:
It is crucial to distinguish the two types of cyclists (subscribers and occasional riders) who use bicycles at different times of the day. We can assume a few factors, one of which is that subscribers may use bicycles during their routine daily activities, such as going to work (between 5 and 8 a.m. on a weekday) and returning from work (between 4 and 6 p.m. midweek).

ds_all_trips_clean_length_correct %>%
group_by(rideable_type) %>%
summarise(count = length(ride_id),
'%' = (length(ride_id) / nrow(ds_all_trips_clean_length_correct)) * 100,
'members_p' = (sum(member_casual == "member") / length(ride_id)) * 100,
'casual_p' = (sum(member_casual == "casual") / length(ride_id)) * 100,
'member_casual_perc_difer' = members_p - casual_p)

ggplot(ds_all_trips_clean_length_correct, aes(rideable_type, fill=member_casual)) +
guides(fill = guide_legend(title = "User category")) +
scale_fill_discrete(labels = c("Casuals", "Members")) +
scale_y_continuous("number of runs", labels = scales::comma) +
labs(x="Bicycle type", title="Chart 7 - Distribution by type of bicycle") +
geom_bar()

It is important to note that:

• “docked” bicycles have a low volume of rides and are used only by occasional users;
• subscribers have a greater preference for classic bicycles, 66 per cent compared to 34 per cent of occasional users;
• Even for e-bikes, subscribers have a greater preference (57 per cent) than occasional subscribers (43 per cent).

Knowing which stations are the most popular for marketing efforts will be helpful. To find out which ones they are, we will use the name of the departure and arrival stations and count the number of trips that start or end on them.

Departure Stations —> Sort starting with those with the highest number of runs:

ds_all_trips_clean_length_correct %>% 
 group_by(start_station_name) %>%
 summarise(
 count = length(ride_id), 
 #ride_id = n(),
 ) %>%
 slice_max(count, n = 10)

Departure Stations —> those most used by subscribers:

ds_all_trips_clean_length_correct %>% 
 filter(member_casual=='member') %>%
 group_by(start_station_name) %>%
 summarise(count = length(ride_id)) %>%
 arrange(desc(count), desc(start_station_name)) %>% 
 slice_max(count, n = 10)

Departure Stations —> those most used by the occasional:

ds_all_trips_clean_length_correct %>% 
 filter(member_casual=='casual') %>%
 group_by(start_station_name) %>%
 summarise(count = length(ride_id)) %>%
 arrange(desc(count), desc(start_station_name)) %>% 
 slice_max(count, n = 10)

Departure Stations —> we now compare the usage of the starting stations with the two types of users (subscribers and occasional users) to check the percentages of use and the time of highest frequency, and we save it all in a dedicated dataset so that we can use it for all filtering and visualization operations:

colnames(ds_all_trips_clean_length_correct)

ds_all_trips_clean_length_correct_station_start <- ds_all_trips_clean_length_correct %>%
 group_by(start_station_name) %>% 
 summarise(count = length(ride_id),
 '%' = (length(ride_id) / nrow(ds_all_trips_clean_length_correct)) * 100,
 'members_p' = (sum(member_casual == "member") / length(ride_id)) * 100,
 'casual_p' = (sum(member_casual == "casual") / length(ride_id)) * 100,
 'member_casual_perc_difer' = members_p - casual_p,
 'members' = (sum(member_casual == "member")),
 'casual' = (sum(member_casual == "casual")),
 'member_casual_difer' = members - casual,
 'hours_first_race' = min(start_hour, na.rm = TRUE),
 'hours_last_race' = max(start_hour, na.rm = TRUE),
 'data_ds_start_at' = min(started_at, na.rm = TRUE),
 'data_ds_end_at' = max(ended_at, na.rm = TRUE),
 'hours_mean' = mean(start_hour, na.rm = TRUE),
 'hours_median' = median(start_hour, na.rm = TRUE)) %>% 
 arrange(desc(count), desc(start_station_name))
 #slice(1:10) 
ds_all_trips_clean_length_correct_station_start

Arrival stations —> Sort starting with those with the highest number of runs:

ds_all_trips_clean_length_correct %>% 
 group_by(end_station_name) %>%
 summarise(
 count = length(ride_id), 
 #ride_id = n(),
 ) %>%
 slice_max(count, n = 10)

Arrival stations —>those most used by subscribers:

ds_all_trips_clean_length_correct %>% 
 filter(member_casual=='member') %>%
 group_by(end_station_name) %>%
 summarise(count = length(ride_id)) %>%
 arrange(desc(count), desc(end_station_name)) %>% 
 slice_max(count, n = 10)

Arrival stations —>those most used by the occasional:

ds_all_trips_clean_length_correct %>% 
 filter(member_casual=='casual') %>%
 group_by(end_station_name) %>%
 summarise(count = length(ride_id)) %>%
 arrange(desc(count), desc(end_station_name)) %>% 
 slice_max(count, n = 10)

Arrival stations —> We now compare the use of the arrival stations with the two types of users (subscribers and occasional users) to check the percentages of use and the time of highest frequency, and we save it all in a dedicated dataset so that we can use it for all filtering and visualization operations:

ds_all_trips_clean_length_correct_station_end <- ds_all_trips_clean_length_correct %>%
 group_by(end_station_name) %>%
 summarise(count = length(ride_id),
 '%' = (length(ride_id) / nrow(ds_all_trips_clean_length_correct)) * 100,
 'members_p' = (sum(member_casual == "member") / length(ride_id)) * 100,
 'casual_p' = (sum(member_casual == "casual") / length(ride_id)) * 100,
 'member_casual_perc_difer' = members_p - casual_p,
 'members' = (sum(member_casual == "member")),
 'casual' = (sum(member_casual == "casual")),
 'member_casual_difer' = members - casual,
 'hours_first_race' = min(start_hour, na.rm = TRUE),
 'hours_last_race' = max(start_hour, na.rm = TRUE),
 'data_ds_start_at' = min(started_at, na.rm = TRUE),
 'data_ds_end_at' = max(ended_at, na.rm = TRUE),
 'hours_mean' = mean(start_hour, na.rm = TRUE),
 'hours_median' = median(start_hour, na.rm = TRUE)) %>% 
 arrange(desc(count), desc(end_station_name))
 #slice(1:10) 
ds_all_trips_clean_length_correct_station_end

Departure/arrival station considerations:

It can be seen that the most frequently used stations are different depending on the type of user. This may be related to the type of bicycle use; as occasional users’ use is more leisure-related, their stations may be closer to leisure places, while subscribers use bicycles mainly to commute to work.

The most significant traffic, both departing and arriving, is concentrated at the “Streeter Dr & Grand Ave” station, where, however, 77 per cent of leaving and 80 per cent of coming users belong to the “occasional” category, with the average peak hour (every week) around 2 p.m.

On the other hand, traffic for annual subscribers is concentrated on the “Kingsbury St & Kinzie St” station, with about 75 per cent outbound and 77 per cent inbound, compared to the occasional subscribers who frequent this station. At this station, on average (every week), the peak time is around 1 pm.

The stations most used by subscribers are:
“Kingsbury St & Kinzie St”, “Clark St & Elm St”, and “Wells St & Concord Ln”.

The stations most used by casual users are:
“Streeter Dr & Grand Ave”, “DuSable Lake Shore Dr & Monroe St”, and “Millennium Park”.

Now let’s take a look at some variables in the data set.

We first display some summary statistics from the data set.

summary(ds_all_trips_clean_length_correct$ride_length_mins)

Minimum and maximum values can be a problem for plotting some graphs. Why do some bicycles’ travel times have such a high maximum value? Perhaps there is some malfunction of the stations returning wrong dates.

We first divide the population into several equal parts to check in which range the anomaly (ventili) is present:

ventiles = quantile(ds_all_trips_clean_length_correct$ride_length_mins, seq(0, 1, by=0.05))
ventiles

We can see that:

• the difference between 0% and 100% is 34354.07 minutes (over 572 hours, almost 24 days);
• the difference between 0% and 95% is 47.5 minutes;
• values below 5% are too short to be considered a real run (less than 3 minutes).

Let’s look at which rides have the highest travel times to see if we can find any references that might provide valuable insights into the reason for this abnormal use:

ds_all_trips_clean_length_correct %>% arrange(desc(ride_length_mins))

Instead, here are the runs listed from the minimum duration:

ds_all_trips_clean_length_correct %>% arrange((ride_length_mins))

Checking the various entries (e.g., departure and arrival stations) would seem to be no problem; however, we note that all the outliers for rides with high times are confined to “docked” bikes.
More information about this data will be requested from relevant stakeholders to see if the company considers it normal.

We also check by filtering for the range > 95%:

ds_all_trips_clean_length_correct_outliners_max <- ds_all_trips_clean_length_correct %>%
 filter(ride_length_mins > as.numeric(ventiles['95%'])) %>%
 arrange(desc(ride_length_mins))
ds_all_trips_clean_length_correct_outliners_max

Once the appropriate checks have been made, delete the last dataset just created (…_outliners_max):

rm(ds_all_trips_clean_length_correct_outliners_max)

Attention: the analysis continues by creating a dataset without outliners (“…_outliners_without“), with a subset of data that excludes the outlier minimum/maximum extremes identified earlier, then with an optimal range between 5 and 95 per cent.
This choice is also dictated by the need to perform verifications that would be difficult to interpret in the graphs with the presence of these outliers.

ds_all_trips_clean_length_correct_outliners_without <- ds_all_trips_clean_length_correct %>% 
 filter(ride_length_mins > as.numeric(ventiles['5%'])) %>%
 filter(ride_length_mins < as.numeric(ventiles['95%']))
print(paste("Removed", nrow(ds_all_trips_clean_length_correct) - nrow(ds_all_trips_clean_length_correct_outliners_without), "rows as outliner" ))

A box plot is one of the first interactions between the columns and the running time, with underlying schemes based on the casual_members column. Also, for summarized data.

ds_all_trips_clean_length_correct_outliners_without %>% 
 group_by(member_casual) %>% 
 summarise(mean = mean(ride_length_mins),
 'first_quarter' = as.numeric(quantile(ride_length_mins, .25)),
 'median' = median(ride_length_mins),
 'third_quarter' = as.numeric(quantile(ride_length_mins, .75)),
 'IR' = third_quarter - first_quarter)

ggplot(ds_all_trips_clean_length_correct_outliners_without, aes(x=member_casual, y=ride_length_mins, fill=member_casual)) +
 guides(fill = guide_legend(title = "User category")) +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 labs(x="Members vs Casuals", y="Travelling time", title="Chart 8 - Distribution of travel time for Casuals/Members") +
 geom_boxplot()

It is important to note that:

• occasional riders have more time to ride than subscribers;
• The mean and IQR are also larger for occasional clients.

Let’s see if finding other helpful information when processing the day of the week is possible.

ggplot(ds_all_trips_clean_length_correct_outliners_without, aes(x=weekday, y=ride_length_mins, fill=member_casual)) +
 geom_boxplot() +
 facet_wrap(~ member_casual) +
 guides(fill = guide_legend(title = "User category")) +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 labs(x="Day", y="Travelling time", title="Diagram 9 var.a - Distribution of rental time by day of the week") +
 theme(axis.text.x=element_text(angle=45,hjust=1))

Some considerations:

• pedalling time for subscribers remains unchanged during midweek, while it increases on weekends;
• casuals follow a more curved distribution, with a peak on Saturday/Sunday and the lowest values on Tuesday/Wednesday/Thursday.

Finally, we elaborate according to bicycle type.

ggplot(ds_all_trips_clean_length_correct_outliners_without, aes(x=rideable_type, y=ride_length_mins, fill=member_casual)) +
 geom_boxplot() +
 facet_wrap(~ member_casual) +
 guides(fill = guide_legend(title = "User category")) +
 scale_fill_discrete(labels = c("Casuals", "Members")) +
 labs(x="Bicycle type", y="Travelling time", title="Graph 10 - Distribution of rental time by type of bicycle")
 #coord_flip()

Some considerations:

• electric bikes have a lower riding time than other bikes, both for subscribers and occasional riders;
• “docked” bicycles have more travel time (data that would be “out of scale” in the graph if we used the dataset including the outliners removed above, referring to the > 95% interval) and are used only by occasional customers.