Cyclistic is fictional company that running a bike-sharing business, located at Chicago, United State. The dataset is created based on a real company named Divvy. Cyclistic dataset stored in AWS cloud store in ZIP format, and updated monthly from January,2014 till May,2022. This dataset is collected by Divvy company itself, and maintained by Manucipal of Chicago.
For analysis purpose, i deriberately take only 1 year data from May, 2022 until May,2021. This sample contains about 6.000.000 rows of data, and 13 attributes.
There are five (from original six) data analysis phase that i did for this project, ie: 1. Ask 2. Prepare 3. Process 4. Analyze 5. Share
The first phase of analysis is ASK. Fortunately, this capstone project has define what we need to seek as final goal clearly. The business statement of Cyclistic Capstone Project is this following : Find the different of Cyclistic member user and casual user CAUTION! please note that the origional objective that declared by Google for Cyclistic project is to give recommendation that can increase convertion from casual member into annual member. But, my judgement upon the existed data conclude that this dataset insufficient to generate accurate recommendation. For next analysis project, i suggest few data attribute to collect. First, cycling distance to anaylze geospatial cycling behavior between member and casual. Second,Weather to analyze the impact of weather on user decision toward Cyclistic service usage.
For prepare phase, i didn’t do anything because all the data has organized and collected by credible authorization.
The following script, explain each step i took during data cleaning process:
install.packages('tidyverse')
library(tidyverse)
library(vroom)
library(skimr)
# collapse data into 4 parts
cyclis_04052022_df <- vroom(c("202205_bike_tripdata.csv","202204_bike_tripdata.csv"))
cyclis_01032022_df <- vroom(c("202201_bike_tripdata.csv","202202_bike_tripdata.csv","202203_bike_tripdata.csv"))
cyclis_10122021_df <- vroom(c("202110_bike_tripdata.csv","202111_bike_tripdata.csv","202112_bike_tripdata.csv"))
cyclis_07092021_df <- vroom(c("202107_bike_tripdata.csv","202108_bike_tripdata.csv","202109_bike_tripdata.csv"))
cyclis_05062021_df <- vroom(c("202105_bike_tripdata.csv","202106_bike_tripdata.csv"))
# skim all data to identify missing value, format data, and unique value
skim_without_charts(cyclis_04052022_df)
skim_without_charts(cyclis_01032022_df)
skim_without_charts(cyclis_10122021_df)
skim_without_charts(cyclis_07092021_df)
skim_without_charts(cyclis_05062021_df)
# omit all missing
cyclis_05062021_df <- na.omit(cyclis_05062021_df)
cyclis_07092021_df <- na.omit(cyclis_07092021_df)
cyclis_10122021_df <- na.omit(cyclis_10122021_df)
cyclis_01032022_df <- na.omit(cyclis_01032022_df)
cyclis_04052022_df <- na.omit(cyclis_04052022_df)
# export files
write.csv(cyclis_05062021_df, file="cyclis_05062021_df")
write.csv(cyclis_07092021_df, file="cyclis_07092021_df")
write.csv(cyclis_10122021_df, file="cyclis_10122021_df.csv")
write.csv(cyclis_01032022_df, file="cyclis_01032022_df.csv")
write.csv(cyclis_04052022_df, file="cyclis_04052022_df.csv")
# take all lat and lng station
cyclis_station_geo <- cyclis_05062021_df %>% select(start_lat,start_lng,
end_lat,end_lng, member_casual)
cyclis_station_geo <- rbind(cyclis_station_geo, cyclis_07092021_df[9:13])
cyclis_station_geo <- rbind(cyclis_station_geo, cyclis_10122021_df[9:13])
cyclis_station_geo <- rbind(cyclis_station_geo, cyclis_01032022_df[9:13])
cyclis_station_geo <- rbind(cyclis_station_geo, cyclis_04052022_df[9:13])
# Calculate distance
# Load package geosphere
install.packages("geosphere")
library(geosphere)
#use Geo Distance Function
cyclis_station_geo <- cyclis_station_geo %>%
mutate (geo_dist = (distGeo(cbind(start_lng,start_lat),cbind(end_lng,end_lat))))
head(cyclis_station_geo)
write.csv(cyclis_station_geo, file="cyclis_station_geo.csv")
library(ggplot2)
# plotting distance
ggplot(cyclis_station_geo) + geom_boxplot(aes(member_casual, geo_dist), fill = c("blue","red"))
cyclis_station_geo %>% group_by(member_casual) %>% summarize(member_median = format(round(median(geo_dist),3),nsmall=3))
# remove outlier (there is only one outlier in this data, we can suspect it for
# cycle thieve, or GPS error; either way, this value is considered useless)
cyclis_iqr <- IQR(cyclis_station_geo$geo_dist)
cyclis_station_geo <- subset(cyclis_station_geo,cyclis_station_geo$geo_dist > (896.2-1.5 * cyclis_iqr) & cyclis_station_geo$geo_dist < (2793.3 + 1.5*cyclis_iqr))
write.csv(cyclis_station_geo, file="cyclis_station_geo.csv")
head(cyclis_04052022_df)
# Based on datetime data structure, i conclude that cyclis is cross-section data,
# main reason because it doesn't really trying to depict one variable overtime,
# rather this data is just listing all order based on ride_id
# Descriptive Statistic of cyclis_station_geo
summary(cyclis_station_geo)
hist(cyclis_station_geo$start_lat)
hist(cyclis_station_geo$start_lng)
hist(cyclis_station_geo$end_lat)
hist(cyclis_station_geo$end_lng)
hist(cyclis_station_geo$geo_dist)
# There is an indication that the geo_dist data is not normally distributed.
# Proceed, to perform histogram and Kolmogorov-Smirnov test for normality
hist(cyclis_station_geo$geo_dist)
ks.test(cyclis_station_geo$geo_dist,'pnorm')
# These two tests formallu prove that geo_dist is not normally distributed.
# because of its large sample size, t-test still relevant in this data, despite the distribution
geo_dist_casual <- cyclis_station_geo %>% filter(member_casual == 'casual') %>% select(geo_dist)
geo_dist_member <- cyclis_station_geo %>% filter(member_casual == 'member') %>% select(geo_dist)
# Use Welch's Test (becuase equal variance unassumed)
t.test(geo_dist_casual,geo_dist_member, var.equal=FALSE)
t.test(cyclis_station_geo$geo_dist~cyclis_station_geo$member_casual, var.equal=FALSE)The following script define each step i did for analysis phase. For readability, i divide the analysis into two major parts, i.e: 1) Analysis based on cycling displacement, 2) Analysis based on cycling duration
# load data
cyclis_station <- read.csv("cyclis_station_geo.csv")
# load neccesary library
install.packages("tidyverse")
library(tidyverse)
library(ggplot2)
# summarize the data
head(cyclis_station)
str(cyclis_station)
summary(cyclis_station)
cyclis_station <- cyclis_station[-1]
# check all variables variance
for(i in seq_along(cyclis_station)){
if(class(cyclis_station[,i]) %in% c("numeric","integer")){
print(sd(cyclis_station[,i]))
}else{
print("non-numeric")
}
}
# check data distribution with histogram plot using Hmisc package
install.packages("Hmisc")
library(Hmisc)
library(scales)
hist.data.frame(cyclis_station[-5])
#try plot bar to calculate total user
cyclis_station %>% count(member_casual=='casual')
#barplot for total user by user category
ggplot(cyclis_station,aes(member_casual,fill=member_casual))+geom_bar() +
geom_text(aes(label=..count..), stat="count", vjust=1.5, position=position_dodge(.9))+
labs(x="User Category", y="Total User") +
ggtitle("Total Cyclis User (in million)", "as of May,2022")+
scale_y_continuous(labels= label_number(suffix = " M", scale=1e-6))+
theme(legend.position="none")
# check how much zero cycling distance
zero_geo_dist <- cyclis_station %>% filter(geo_dist==0)
head(zero_geo_dist)
summary(zero_geo_dist)
hist.data.frame(zero_geo_dist)
# by looking at the histogram, zero distance cycling only seem to only happen with casual user
# count the user
zero_geo_dist %>% count(member_casual=='casual')
write.csv(zero_geo_dist, file="zero_geo_dist.csv")
# total user by user category in zero cycling distance
ggplot(zero_geo_dist,aes(member_casual,fill=member_casual))+geom_bar() +
geom_text(aes(label=..count..), stat="count", vjust=1.5, position=position_dodge(.9))+
labs(x="User Category", y="Total User") +
ggtitle("Number of Zero Cycling Distance User (in million)", "as of May,2022")+
scale_y_continuous(labels= label_number(suffix = " K", scale=0.001))+
theme(legend.position="none")
# Zero distance cycling represents interesting insight.
# This data show us that there are some users who stop at the same station where they started
# For now, we call this cycling as 'round-back user'
# For non-zero cycling distance, i called it 'travel user'
# I will analyze the user difference based on this type of travelling route
nonzero_geo_dist <- cyclis_station %>% filter(geo_dist !=0)
write.csv(nonzero_geo_dist, file="nonzero_geo_dist.csv")
# try to summarize this non-zero geo distance data
summary(nonzero_geo_dist)
hist.data.frame(nonzero_geo_dist[-5])
# focus on geo_dist variable, we seek to analyze the difference of cycling distance
# between user group
# plot geo_dist with boxplot by member_casual
ggplot(nonzero_geo_dist) + geom_boxplot(aes(member_casual,geo_dist), fill=c('blue','green'))
# try violin plot
ggplot(nonzero_geo_dist, aes(member_casual, geo_dist)) +
geom_violin(aes(fill=member_casual))+
geom_boxplot(width=0.1)
# Based on violin plot, we can see median difference in cycling distance between member and casual user
# From this plot, it safe for us to hypothesize that the average cycling distance of member and casual user is statistically different.
# to prove our hypothesis, we can use T-test
# Let's divide the geo_dist into two user groups
# The reason, because we have enough large sample size, that T-Test is feasible for
# non-normal distributed data
# First, let's create two sample of geo_dist divided by member status
nonzero_member_geo <- nonzero_geo_dist %>% filter(member_casual=='member') %>%
select(geo_dist)
nonzero_casual_geo <- nonzero_geo_dist %>% filter(member_casual=='casual') %>%
select(geo_dist)
t.test(nonzero_member_geo,nonzero_casual_geo, paired = FALSE, var.equal=FALSE)
# the test result is sufficient to reject H0, and that prove our hypothesis that
# there are significant difference on the average of cycling distance between
# member and casual user, where casual user cycling farther than member user
mean(nonzero_member_geo$geo_dist)
mean(nonzero_casual_geo$geo_dist)
# load all data
cyclis_05062021_df <- vroom("cyclis_05062021_df.csv")
cyclis_07092021_df <- vroom("cyclis_07092021_df.csv")
cyclis_10122021_df <- vroom("cyclis_10122021_df.csv")
cyclis_01032022_df <- vroom("cyclis_01032022_df.csv")
cyclis_04052022_df <- vroom("cyclis_04052022_df.csv")
# define all columns
str(cyclis_04052022_df)
# take rideable_type, started_at, ended_at, member_casual
cyclis_durtime <- cyclis_05062021_df[c(3:5,14)]
cyclis_durtime <- rbind(cyclis_durtime, cyclis_07092021_df[c(3:5,14)])
cyclis_durtime <- rbind(cyclis_durtime, cyclis_10122021_df[c(3:5,14)])
cyclis_durtime <- rbind(cyclis_durtime, cyclis_01032022_df[c(3:5,14)])
cyclis_durtime <- rbind(cyclis_durtime, cyclis_04052022_df[c(3:5,14)])
# check for any missing value
sum(is.na(cyclis_durtime))
# write csv with fwrite
library(data.table)
library(vroom)
library(tidyverse)
fwrite(cyclis_durtime, file="cyclis_durtime.csv", sep=",")
cyclis_durtime <- vroom("cyclis_durtime.csv")
head(cyclis_durtime)
str(cyclis_durtime)
# summarize ride type
cyclis_durtime %>% group_by(rideable_type) %>% summarise(count=n()) %>%
mutate(prop=count/sum(count))
# load lubridate
library(lubridate)
# check if all cyling done in same day
sum(date(cyclis_durtime$started_at) == date(cyclis_durtime$ended_at))
# calculate cycling duration
cyclis_durtime <-cyclis_durtime %>% mutate(cyclis_dur = difftime(ended_at,started_at, units ="secs"))
#convert difftime to numeric
cyclis_durtime$cyclis_dur<-as.double.difftime(cyclis_durtime$cyclis_dur)
head(cyclis_durtime)
summary(cyclis_durtime)
# we found negative duration in cycling duration.
# Because it doesn't make sense to have negative duration, we can erase it
# There are 112 ride data with negative duration
cyclis_durtime <- cyclis_durtime %>% filter(cyclis_dur > 0)
summary(cyclis_durtime)
sd(cyclis_durtime$cyclis_dur)
#based on data summary,there is high chance that this duration data has outlier
boxplot(cyclis_durtime$cyclis_dur)
# Find the outliers
durtime_iqr <- IQR(cyclis_durtime$cyclis_dur)
cyclis_durtime_nooutlier <- subset(cyclis_durtime, cyclis_durtime$cyclis_dur > (401-1.5*durtime_iqr) & cyclis_durtime$cyclis_dur < (707+1.5*durtime_iqr))
# Check the first group data after cleaning
summary(cyclis_durtime_nooutlier)
boxplot(cyclis_durtime_nooutlier$cyclis_dur)
hist(cyclis_durtime_nooutlier$cyclis_dur)
first_iqr <- IQR(cyclis_durtime_nooutlier$cyclis_dur)
cyclis_durtime_nooutlier <- subset(cyclis_durtime_nooutlier,
cyclis_durtime_nooutlier$cyclis_dur > (368-1.5*first_iqr) &
cyclis_durtime_nooutlier$cyclis_dur < (1002+1.5*first_iqr))
summary(cyclis_durtime_nooutlier)
boxplot(cyclis_durtime_nooutlier$cyclis_dur)
hist(cyclis_durtime_nooutlier$cyclis_dur)
#Called it 1up997sec group
cyclis_1up997sec_durgroup <- cyclis_durtime_nooutlier
summary(cyclis_1up997sec_durgroup$cyclis_dur)
cyclis_1up1952sec_durgroup <- cyclis_1up997sec_durgroup
fwrite(cyclis_1up1952sec_durgroup, file="cyclis_1up1952sec_durgroup.csv", sep = ",")
ggplot(cyclis_1up1952sec_durgroup, aes(cyclis_dur)) +
geom_histogram() +
ggtitle("Cyclistic Usage Duration Distribution","1-1952 seconds")
#analyze the outlier from first group to see if there is some pattern
cyclis_durtime_outlier <- subset(cyclis_durtime,
cyclis_durtime$cyclis_dur < (401-1.5*durtime_iqr) |
cyclis_durtime$cyclis_dur > (707+1.5*durtime_iqr))
boxplot(cyclis_durtime_outlier$cyclis_dur)
summary(cyclis_durtime_outlier)
second_iqr <- IQR(cyclis_durtime_outlier$cyclis_dur)
cyclis_non_outlier_second <- subset(cyclis_durtime_outlier,
cyclis_durtime_outlier$cyclis_dur>(2383 - 1.5*second_iqr) &
cyclis_durtime_outlier$cyclis_dur<(4497 + 1.5*second_iqr))
#Analyze second durtime group
summary(cyclis_non_outlier_second)
hist(cyclis_non_outlier_second$cyclis_dur)
boxplot(cyclis_non_outlier_second$cyclis_dur)
# there are still lot of outliers in cyclis_outlier
third_iqr <- IQR(cyclis_non_outlier_second$cyclis_dur)
summary(cyclis_non_outlier_second$cyclis_dur)
cyclis_non_outlier_second <- subset(cyclis_non_outlier_second,
cyclis_non_outlier_second$cyclis_dur>(2308 - 1.5*third_iqr) &
cyclis_non_outlier_second$cyclis_dur<(3516 + 1.5*third_iqr))
boxplot(cyclis_non_outlier_second$cyclis_dur)
# we cleaned to optimal state This group will called as cyclis_20315326sec_dur
cyclis_2031up5326sec_durgroup <- cyclis_non_outlier_second
summary(cyclis_2031up5326sec_durgroup$cyclis_dur)
fwrite(cyclis_2031up5326sec_durgroup, file="cyclis_2031up5236sec_durgroup.csv", sep = ",")
ggplot(cyclis_2031up5326sec_durgroup, aes(cyclis_dur))+
geom_histogram(fill="blue")+
ggtitle("Cyclistic Usage Duration Distribution","2031-5236 seconds")
# Now, we further the anaylsis of outlier from second group
cyclis_outlier_from_second <- subset(cyclis_durtime_outlier,
cyclis_durtime_outlier$cyclis_dur<(2383 - 1.5*second_iqr) |
cyclis_durtime_outlier$cyclis_dur>(4497 + 1.5*second_iqr))
summary(cyclis_outlier_from_second$cyclis_dur)
boxplot(cyclis_outlier_from_second$cyclis_dur)
# We clean up this group
fourth_iqr <- IQR(cyclis_outlier_from_second$cyclis_dur)
cyclis_non_outlier_third <- subset(cyclis_outlier_from_second,
cyclis_outlier_from_second$cyclis_dur>(8472 - 1.5*fifth_iqr) &
cyclis_outlier_from_second$cyclis_dur<(11448 + 1.5*fourth_iqr))
summary(cyclis_non_outlier_third$cyclis_dur)
boxplot(cyclis_non_outlier_third$cyclis_dur)
fifth_iqr <- IQR(cyclis_non_outlier_third$cyclis_dur)
cyclis_non_outlier_third <- subset(cyclis_non_outlier_third,
cyclis_non_outlier_third$cyclis_dur>(8365 - 1.5*fifth_iqr) &
cyclis_non_outlier_third$cyclis_dur<(10668 + 1.5*fifth_iqr))
# Data cleaned. We called it cyclis_8365up10668sec_durtime
cyclis_8365up10668sec_durtime <- cyclis_non_outlier_third
summary(cyclis_8365up10668sec_durtime$cyclis_dur)
cyclis_8365up14120sec_durtime <- cyclis_8365up10668sec_durtime
ggplot(cyclis_8365up14120sec_durtime, aes(cyclis_dur))+
geom_histogram(fill="green")+
ggtitle("Cyclistic Usage Duration Distribution","2H19M - 3H55M")
fwrite(cyclis_8365up14120sec_durtime, file="cyclis_8365up14120sec_durtime.csv", sep = ",")
# Split for the outlier from third group
cyclis_outlier_from_third <- subset(cyclis_outlier_from_second,
cyclis_outlier_from_second$cyclis_dur<(8634 - 1.5*fourth_iqr) |
cyclis_outlier_from_second$cyclis_dur>(13772 + 1.5*fourth_iqr))
#analyze outlier from thir group
boxplot(cyclis_outlier_from_third$cyclis_dur)
#clean up the outlier from third
summary(cyclis_outlier_from_third$cyclis_dur)
sixth_iqr <- IQR(cyclis_outlier_from_third$cyclis_dur)
cyclis_non_outlier_fourth <- subset(cyclis_outlier_from_third,
cyclis_outlier_from_third$cyclis_dur>(30362 - 1.5*sixth_iqr)&
cyclis_outlier_from_third$cyclis_dur<(71744 + 1.5*sixth_iqr))
boxplot(cyclis_non_outlier_fourth$cyclis_dur)
summary(cyclis_non_outlier_fourth$cyclis_dur)
seventh_iqr <- IQR(cyclis_non_outlier_fourth$cyclis_dur)
cyclis_non_outlier_fourth <- subset(cyclis_non_outlier_fourth,
cyclis_non_outlier_fourth$cyclis_dur>(29023 - 1.5*seventh_iqr)&
cyclis_non_outlier_fourth$cyclis_dur<(60561 + 1.5*seventh_iqr))
hist(cyclis_non_outlier_fourth$cyclis_dur)
# this group will called as cyclis_21490up107722sec_durtime
cyclis_21490up107722sec_durtime <- cyclis_non_outlier_fourth
summary(cyclis_21490up107722sec_durtime$cyclis_dur)
ggplot(cyclis_21490up107722sec_durtime, aes(cyclis_dur))+
geom_histogram(fill="purple")+
ggtitle("Cyclis Usage Duration Distribution","5H58M-29H55M")
fwrite(cyclis_21490up107722sec_durtime,file="cyclis_21490up107722sec_durtime.csv",sep = ",")
#split the outlier from fourth
cyclis_outlier_from_fourth <- subset(cyclis_outlier_from_third,
cyclis_outlier_from_third$cyclis_dur<(30362 - 1.5*sixth_iqr)|
cyclis_outlier_from_third$cyclis_dur>(71744 + 1.5*sixth_iqr))
# ini adalah grup terakhir.
# further analysis
user_cat_1up1952sec<- cyclis_1up1952sec_durgroup %>% group_by(member_casual) %>%
summarise(total_user=n(), user_percentage=total_user/nrow(cyclis_1up1952sec_durgroup))
user_cat_1up1952sec
# member_casual total_user user_percentage
#<chr> <int> <dbl>
# 1 casual 1736684 0.389
# 2 member 2726043 0.611
user_cat_2031up5236sec<- cyclis_2031up5236sec_durgroup %>%
group_by(member_casual) %>%
summarise(total_user=n(),user_percentage=total_user/nrow(cyclis_2031up5236sec_durgroup))
user_cat_2031up5236sec
## A tibble: 2 x 3
#member_casual total_user user_percentage
#<chr> <int> <dbl>
# 1 casual 370297 0.738
# 2 member 131513 0.262
user_cat_8365up14120sec<-cyclis_8365up14120sec_durtime %>%
group_by(member_casual) %>%
summarise(total_user=n(), user_percentage=total_user/nrow(cyclis_8365up14120sec_durtime))
user_cat_8365up14120sec
# member_casual total_user user_percentage
# <chr> <int> <dbl>
# 1 casual 36005 0.943
# 2 member 2162 0.0566
user_cat_21490up107722sec<- cyclis_21490up107722sec_durtime %>%
group_by(member_casual) %>%
summarise(total_user=n(), user_percentage=total_user/nrow(cyclis_21490up107722sec_durtime))
user_cat_21490up107722sec
# member_casual total_user user_percentage
# <chr> <int> <dbl>
# 1 casual 5483 0.873
# 2 member 800 0.127
# load data after error
cyclis_1up1952sec_durgroup <- vroom("cyclis_1up1952sec_durgroup.csv")
cyclis_2031up5236sec_durgroup <- vroom("cyclis_2031up5236sec_durgroup.csv")
cyclis_8365up14120sec_durtime <- vroom("cyclis_8365up14120sec_durtime.csv")
cyclis_21490up107722sec_durtime <- vroom("cyclis_21490up107722sec_durtime.csv")
# Visualize
ggplot(user_cat_1up1952sec, aes(x="",y=user_percentage, fill=member_casual))+
geom_col(color="black")+
geom_label(aes(label=paste0(round(user_percentage,2)*100,"%")),
position=position_stack(vjust=0.5),
show.legend = FALSE)+
coord_polar(theta="y")+
scale_fill_brewer()
ggplot(user_cat_2031up5236sec, aes(x="",y=user_percentage, fill=member_casual))+
geom_col(color="black")+
geom_label(aes(label=paste0(round(user_percentage,2)*100,"%")),
position=position_stack(vjust=0.5),
show.legend = FALSE)+
coord_polar(theta="y")+
scale_fill_brewer()
ggplot(user_cat_21490up107722sec, aes(x="",y=user_percentage, fill=member_casual))+
geom_col(color="black")+
geom_label(aes(label=paste0(round(user_percentage,2)*100,"%")),
position=position_stack(vjust=0.5),
show.legend = FALSE)+
coord_polar(theta="y")+
scale_fill_brewer()
ggplot(user_cat_8365up14120sec, aes(x="",y=user_percentage, fill=member_casual))+
geom_col(color="black")+
geom_label(aes(label=paste0(round(user_percentage,2)*100,"%")),
position=position_stack(vjust=0.5),
show.legend = FALSE)+
coord_polar(theta="y")+
scale_fill_brewer()
# Analyze by day
by_day_8365up14120sec <- cyclis_8365up14120sec_durtime %>%
group_by(date(started_at),member_casual) %>%
summarise(user_by_day = n(), user_by_log=(log(n())))
head(by_day_8365up14120sec)
colnames(by_day_8365up14120sec)<-c('date','user_cat','user_by_day','user_by_log')
ggplot(by_day_8365up14120sec, aes(date,user_by_log,color=user_cat))+
geom_point()
ggplot(by_day_8365up14120sec, aes(date,user_by_log,color=user_cat))+
geom_smooth(method = "gam")
fwrite(by_day_8365up14120sec, file="by_day_8365up14120.csv", sep=",")
by_day_1up1952sec <- cyclis_1up1952sec_durgroup %>%
group_by(date(started_at),member_casual) %>%
summarise(user_by_day = n(), user_by_log=(log(n())))
head(by_day_1up1952sec)
colnames(by_day_1up1952sec)<-c('date','user_cat','user_by_day','user_by_log')
ggplot(by_day_1up1952sec, aes(date,user_by_log,color=user_cat))+
geom_point()
ggplot(by_day_1up1952sec, aes(date,user_by_log,color=user_cat))+
geom_smooth(method = "gam")
fwrite(by_day_1up1952sec, file="by_day_1up1952sec.csv",sep=",")
by_day_2031up5236sec <- cyclis_2031up5236sec_durgroup %>%
group_by(date(started_at),member_casual) %>%
summarise(user_by_day=n(), user_by_log=log(n()))
head(by_day_2031up5236sec)
colnames(by_day_2031up5236sec)<-c('date','user_cat','user_by_day','user_by_log')
ggplot(by_day_2031up5236sec, aes(date,user_by_log,color=user_cat))+
geom_point()
ggplot(by_day_2031up5236sec, aes(date,user_by_log,color=user_cat))+
geom_smooth(method="gam")
fwrite(by_day_2031up5236sec, file="by_day_2031up5236sec.csv", sep=",")
by_day_21490up107722sec <- cyclis_21490up107722sec_durtime %>%
group_by(date(started_at), member_casual) %>%
summarise(user_by_day=n(), user_by_log=(log(n())))
head(by_day_21490up107722sec)
colnames(by_day_21490up107722sec)<-c('date','user_cat','user_by_day','user_by_log')
ggplot(by_day_21490up107722sec, aes(date,user_by_log,color=user_cat))+
geom_point()
ggplot(by_day_21490up107722sec, aes(date,user_by_log))+
geom_line(alpha=0.8)+
geom_smooth(method="gam")
fwrite(by_day_21490up107722sec, file="by_day_21490up107722sec.csv",sep=",")
# analysis by type of bike
# to imporve ggplot, use package scales
library(scales)
by_type_1up1952sec<-cyclis_1up1952sec_durgroup %>%
group_by(member_casual, rideable_type) %>%
summarise(total_user=n())
head(by_type_1up1952sec)
by_type_1up1952sec$member_casual <- as_factor(by_type_1up1952sec$member_casual)
ggplot(by_type_1up1952sec, aes(rideable_type, total_user))+
geom_col(aes(fill=member_casual))+
ggtitle("Cyclistic Usage By Bike Type (in thousand)","1 - 1952 seconds")+
geom_text(aes(label=total_user), color="white", position=position_stack(vjust=0.6))+
scale_y_continuous(labels = label_number(scale=0.001))+
theme_classic()
by_type_2031up5236sec <- cyclis_2031up5236sec_durgroup %>%
group_by(member_casual, rideable_type) %>%
summarise(total_user=n())
ggplot(by_type_2031up5236sec, aes(rideable_type, total_user))+
geom_col(aes(fill=member_casual))+
ggtitle("Cyclistic Usage By Bike Type (in thousand","2031 - 5236 seconds")+
geom_text(aes(label=total_user), color="white",position=position_stack(vjust=0.5))+
scale_y_continuous(labels = label_number(scale=0.001))+
theme_classic()
by_type_21490up107722sec <- cyclis_21490up107722sec_durtime %>%
group_by(member_casual, rideable_type) %>%
summarise(total_user=n())
ggplot(by_type_21490up107722sec, aes(rideable_type, total_user))+
geom_col(aes(fill=member_casual))+
ggtitle("Cyclistic Usage By Bike Type (in thousand)","21490 - 107722 seconds")+
geom_text(aes(label=total_user), color="black",position=position_jitter())+
scale_y_continuous(labels = label_number(scale=0.001))+
theme_classic()
by_type_8365up14120sec <- cyclis_8365up14120sec_durtime %>%
group_by(member_casual, rideable_type) %>%
summarise(total_user=n())
ggplot(by_type_8365up14120sec, aes(rideable_type, total_user))+
geom_col(aes(fill=member_casual))+
ggtitle("Cyclistic Usage By Bike Type (in thousand)","8365 - 14120 seconds")+
geom_text(aes(label=total_user), color="black",position=position_stack(vjust=0.1))+
scale_y_continuous(labels = label_number(scale=0.001))+
theme_classic()
# use kolmogrov-smirnov to test if each group really different or not
install.packages("dgof")
library(dgof)
# test 1up1920sec vs 2031up5236sec
ks.test(cyclis_1up1952sec_durgroup$cyclis_dur,cyclis_2031up5236sec_durgroup$cyclis_dur, exact=FALSE)
# Two-sample Kolmogorov-Smirnov test
#data: cyclis_1up1952sec_durgroup$cyclis_dur and cyclis_2031up5236sec_durgroup$cyclis_dur
#D = 1, p-value < 2.2e-16
# test 1up1920sec vs 8365up14120sec
ks.test(cyclis_1up1952sec_durgroup$cyclis_dur, cyclis_8365up14120sec_durtime$cyclis_dur, exact = FALSE)
#data: cyclis_1up1952sec_durgroup$cyclis_dur and cyclis_8365up14120sec_durtime$cyclis_dur
#D = 1, p-value < 2.2e-16
#alternative hypothesis: two-sided
# test 1up1920sec vs 21490up107722sec
ks.test(cyclis_1up1952sec_durgroup$cyclis_dur, cyclis_21490up107722sec_durtime$cyclis_dur)
#Two-sample Kolmogorov-Smirnov test
#data: cyclis_1up1952sec_durgroup$cyclis_dur and cyclis_21490up107722sec_durtime$cyclis_dur
#D = 1, p-value < 2.2e-16
#alternative hypothesis: two-sided
# test 2031up5236sec vs 8365up14120sec
ks.test(cyclis_2031up5236sec_durgroup$cyclis_dur,cyclis_8365up14120sec_durtime$cyclis_dur)
# Two-sample Kolmogorov-Smirnov test
#data: cyclis_2031up5236sec_durgroup$cyclis_dur and cyclis_8365up14120sec_durtime$cyclis_dur
#D = 1, p-value < 2.2e-16
#alternative hypothesis: two-sided
# test 2031up5236sec vs 21490up107722sec
ks.test(cyclis_2031up5236sec_durgroup$cyclis_dur,cyclis_21490up107722sec_durtime$cyclis_dur)
# Two-sample Kolmogorov-Smirnov test
#data: cyclis_2031up5236sec_durgroup$cyclis_dur and cyclis_21490up107722sec_durtime$cyclis_dur
#D = 1, p-value < 2.2e-16
#alternative hypothesis: two-sided
# test 8365up14120sec vs 21490up107722sec
ks.test(cyclis_8365up14120sec_durtime$cyclis_dur,cyclis_21490up107722sec_durtime$cyclis_dur)
# Two-sample Kolmogorov-Smirnov test
#data: cyclis_8365up14120sec_durtime$cyclis_dur and cyclis_21490up107722sec_durtime$cyclis_dur
#D = 1, p-value < 2.2e-16
#alternative hypothesis: two-sided
# each group indeed has different distribution
plot(ecdf(cyclis_1up1952sec_durgroup$cyclis_dur),
xlim=range(c(cyclis_1up1952sec_durgroup$cyclis_dur,cyclis_2031up5236sec_durgroup$cyclis_dur)),
col="blue")
plot(ecdf(cyclis_2031up5236sec_durgroup$cyclis_dur),
add=TRUE,
lty="dashed",
col="red")
plot(ecdf(cyclis_1up1952sec_durgroup$cyclis_dur),
xlim=range(c(cyclis_1up1952sec_durgroup$cyclis_dur,cyclis_8365up14120sec_durtime$cyclis_dur)),
col="blue")
plot(ecdf(cyclis_8365up14120sec_durtime$cyclis_dur),
add=TRUE,
lty="dashed",
col="red")
plot(ecdf(cyclis_1up1952sec_durgroup$cyclis_dur),
xlim=range(c(cyclis_1up1952sec_durgroup$cyclis_dur,cyclis_21490up107722sec_durtime$cyclis_dur)),
col="blue")
plot(ecdf(cyclis_21490up107722sec_durtime$cyclis_dur),
add=TRUE,
lty="dashed",
col="red")
plot(ecdf(cyclis_2031up5236sec_durgroup$cyclis_dur),
xlim=range(c(cyclis_2031up5236sec_durgroup$cyclis_dur,cyclis_8365up14120sec_durtime$cyclis_dur)),
col="blue")
plot(ecdf(cyclis_8365up14120sec_durtime$cyclis_dur),
add=TRUE,
lty="dashed",
col="red")
plot(ecdf(cyclis_2031up5236sec_durgroup$cyclis_dur),
xlim=range(c(cyclis_2031up5236sec_durgroup$cyclis_dur,cyclis_21490up107722sec_durtime$cyclis_dur)),
col="blue")
plot(ecdf(cyclis_21490up107722sec_durtime$cyclis_dur),
add=TRUE,
lty="dashed",
col="red")
plot(ecdf(cyclis_8365up14120sec_durtime$cyclis_dur),
xlim=range(c(cyclis_8365up14120sec_durtime$cyclis_dur,cyclis_21490up107722sec_durtime$cyclis_dur)),
col="blue")
plot(ecdf(cyclis_21490up107722sec_durtime$cyclis_dur),
add=TRUE,
lty="dashed",
col="red")
by_type_1up1952sec
by_type_2031up5236sec
by_type_8365up14120sec
by_type_21490up107722sec
head(cyclis_1up1952sec_durgroup)
# analyze by time user
head(cyclis_8365up14120sec_durtime)
by_time_8365up14120sec <- cyclis_8365up14120sec_durtime %>%
group_by(year(started_at),month(started_at,label=TRUE),day(started_at),
wday(started_at, label=TRUE),member_casual) %>%
summarise(total_user=n())
by_mean_time_8365up14120sec <- by_time_8365up14120sec %>%
group_by(day_of_week,cat_user) %>% summarise(average=mean(total_user))
by_mean_time_8365up14120sec
colnames(by_time_8365up14120sec)<-c("year","month","day_of_month","day_of_week","cat_user","total_user")
ggplot(by_mean_time_8365up14120sec, aes(day_of_week,average,group=cat_user))+
geom_col(aes(fill=cat_user),position="dodge")+
geom_text(aes(y=average,label=round(average)),vjust=0,hjust=0.8,position="nudge", color="white")+
ggtitle("Cyclistic User Average by Week Days","8365 - 14120 second")+
theme_dark()
by_time_1up1952sec <- cyclis_1up1952sec_durgroup %>%
group_by(year(started_at),month(started_at,label=TRUE),day(started_at),
wday(started_at, label=TRUE),member_casual) %>%
summarise(total_user=n())
by_mean_time_1up1952sec <- by_time_1up1952sec %>%
group_by(day_of_week,cat_user) %>% summarise(average=mean(total_user))
colnames(by_time_1up1952sec)<-c("year","month","day_of_month","day_of_week","cat_user","total_user")
ggplot(by_mean_time_1up1952sec, aes(day_of_week,average,group=cat_user))+
geom_col(aes(fill=cat_user))+
geom_text(aes(y=average+3,label=round(average)),vjust=-1,position=position_stack(0), color="white",size=4)+
ggtitle("Cyclistic User Average by Week Days","1 - 1952 second")+
theme_dark()+
theme(legend.position = "right")
by_time_2031up5236sec <- cyclis_2031up5236sec_durgroup %>%
group_by(year(started_at),month(started_at,label=TRUE),day(started_at),
wday(started_at, label=TRUE),member_casual) %>%
summarise(total_user=n())
colnames(by_time_2031up5236sec)<-c("year","month","day_of_month","day_of_week","cat_user","total_user")
head(by_time_2031up5236sec)
by_mean_time_2031up5236sec <- by_time_2031up5236sec %>%
group_by(day_of_week,cat_user) %>% summarise(average=mean(total_user))
ggplot(by_mean_time_2031up5236sec, aes(day_of_week,average,group=cat_user))+
geom_col(aes(fill=cat_user), position = "dodge")+
geom_text(aes(y=average,label=round(average)),vjust=0,position=position_nudge(0), color="white",size=4)+
ggtitle("Cyclistic User Average by Week Days","2031 - 5236 second")+
theme_dark()+
theme(legend.position = "right")
by_time_21490up107722sec <- cyclis_21490up107722sec_durtime %>%
group_by(year(started_at),month(started_at,label=TRUE),day(started_at),
wday(started_at, label=TRUE),member_casual) %>%
summarise(total_user=n())
colnames(by_time_21490up107722sec)<-c("year","month","day_of_month","day_of_week","cat_user","total_user")
head(by_mean_time_21490up107722sec)
by_mean_time_21490up107722sec <- by_time_21490up107722sec %>%
group_by(day_of_week,cat_user) %>% summarise(average=mean(total_user))
ggplot(by_mean_time_21490up107722sec, aes(day_of_week,average,group=cat_user))+
geom_col(aes(fill=cat_user), position = "dodge")+
geom_text(aes(y=average,label=round(average)),vjust=0,position=position_nudge(0), color="white",size=4)+
ggtitle("Cyclistic User Average by Week Days","21490 - 107722 second")+
theme_dark()+
theme(legend.position = "right")
# Analsis of cyclistic user by time
by_clock_8365up14120sec <-cyclis_8365up14120sec_durtime %>%
group_by(format(as.POSIXct(cyclis_8365up14120sec_durtime$started_at),format="%H")) %>%
summarise(member_user=sum(member_casual=='member'),casual_user=sum(member_casual=='casual'))
head(cyclis_8365up14120sec_durtime)
colnames(cyclis_8365up14120sec_durtime)<- c('rideable_type','started_at','ended_at','member_casual','cyclis_dur')
head(by_clock_8365up14120sec)
colnames(by_clock_8365up14120sec)<- c("time","member_user","casual_user")
install.packages("patchwork")
library(patchwork)
member_clock_8365 <- ggplot(by_clock_8365up14120sec)+
geom_col(aes(x=time,y=member_user),fill="blue")
casual_clock_8365 <- ggplot(by_clock_8365up14120sec)+
geom_col(aes(time,casual_user), fill="magenta")
by_clock_plot_8365up14120sec <- member_clock_8365+casual_clock_8365
by_clock_1up1952sec <-cyclis_1up1952sec_durgroup %>%
group_by(format(as.POSIXct(cyclis_1up1952sec_durgroup$started_at),format="%H")) %>%
summarise(member_user=sum(member_casual=='member'),casual_user=sum(member_casual=='casual'))
head(by_clock_1up1952sec)
colnames(by_clock_1up1952sec)<- c("time","member_user","casual_user")
member_1up1952sec <- ggplot(by_clock_1up1952sec, aes(x=time))+
geom_col(aes(y=member_user),fill="blue")
casual_1up1952sec <- ggplot(by_clock_1up1952sec, aes(x=time))+
geom_col(aes(y=casual_user),fill="magenta")
member_1up1952sec + casual_1up1952sec
by_clock_2031up523sec <-cyclis_2031up5236sec_durgroup %>%
group_by(format(as.POSIXct(cyclis_2031up5236sec_durgroup$started_at),format="%H")) %>%
summarise(member_user=sum(member_casual=='member'),casual_user=sum(member_casual=='casual'))
head(by_clock_2031up523sec)
colnames(by_clock_2031up523sec)<- c("time","member_user","casual_user")
member_2031up523sec <- ggplot(by_clock_2031up523sec, aes(x=time))+
geom_col(aes(y=member_user),fill="blue")
casual_2031up523sec <- ggplot(by_clock_2031up523sec, aes(x=time))+
geom_col(aes(y=casual_user),fill="magenta")
by_clock_21490up107722sec <-cyclis_21490up107722sec_durtime %>%
group_by(format(as.POSIXct(cyclis_21490up107722sec_durtime$started_at),format="%H")) %>%
summarise(member_user=sum(member_casual=='member'),casual_user=sum(member_casual=='casual'))
head(by_clock_21490up107722sec)
colnames(by_clock_21490up107722sec)<- c("time","member_user","casual_user")
member_21490up107722 <- ggplot(by_clock_21490up107722sec, aes(x=time))+
geom_col(aes(y=member_user),fill="blue")
casual_21490up107722 <- ggplot(by_clock_21490up107722sec, aes(x=time))+
geom_col(aes(y=casual_user),fill="magenta")
by_clock_plot_21490up107722 <- member_21490up107722+casual_21490up107722
by_clock_plot_1up1952 <- member_1up1952sec+casual_1up1952sec
by_clock_plot_2031up5236 <- member_2031up523sec+casual_2031up523sec
by_clock_plot_8365up14120sec