Bike_Maintenance.Rmd

---
title: "Better Management of Bicycle Fleet - BikeShare"
output: 
  powerpoint_presentation:
      reference_doc: Bikes_Template.pptx
df_print: paged
always_allow_html: yes  
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = FALSE)
rm(list=ls(all=TRUE))

if(!require(tidyverse)) install.packages("tidyverse", repos = "http://cran.us.r-project.org")
if(!require(tidymodels)) install.packages("tidymodels", repos = "http://cran.us.r-project.org")
if(!require(lubridate)) install.packages("lubridate", repos = "http://cran.us.r-project.org")
if(!require(ggthemes)) install.packages("ggthemes", repunios = "http://cran.us.r-project.org")
if(!require(patchwork)) install.packages("patchwork", repos = "http://cran.us.r-project.org")
if(!require(gridtext)) install.packages("gridtext", repos = "http://cran.us.r-project.org")
if(!require(grid)) install.packages("grid", repos = "http://cran.us.r-project.org")
if(!require(magick)) install.packages("magick", repos = "http://cran.us.r-project.org")
if(!require(gtools)) install.packages("gtools", repos = "http://cran.us.r-project.org")
if(!require(ggmap)) install.packages("gtools", repos = "http://cran.us.r-project.org")
if(!require(mltools)) install.packages("mltools", repos = "http://cran.us.r-project.org")
if(!require(flextable)) install.packages("knitr", repos = "http://cran.us.r-project.org")
if(!require(officer)) install.packages("knitr", repos = "http://cran.us.r-project.org")


library(magick)
library(tidyverse)
library(tidymodels)
library(ggthemes)
library(patchwork)
library(grid)
library(gridtext)
library(rvest)
library(RColorBrewer)
library(lubridate)
library(flextable)
library(officer)

seasons <- tibble(season=c("DJF","MAM","JJA","SON"),
                  season_name=c("winter","spring","summer","autumn"))
    
```


```{r load_data, echo=FALSE, include=FALSE}
if(file.exists("data.RData")){
    load("data.RData")
}else{
if(file.exists("trip_data.RDS")){
  trip_data <- readRDS("trip_data.RDS")
  
}else{
  library(aws.s3) 
  #filesat https://s3.amazonaws.com/capitalbikeshare-data/index.html"
  bucket_id <- 'capitalbikeshare-data'
  bucket <- get_bucket(bucket_id,max = Inf)
  
  keys <- c()
  for(i in seq(1:length(bucket))){
    keys[i] <- bucket[[i]]$Key
  }
  #ignore 2020 
  keys <- c(keys[grepl('2018',keys)],keys[grepl('2019',keys)],keys[grepl('2017',keys)])
  
  save_object(keys[1], file = keys[1], bucket = bucket_id)
  unzip(keys[1],exdir = "./zip")
  file.remove(keys[1])
  files <- list.files("./zip")
  files <- files[!grepl('MACOSX',files)]
  #first file
  
  trip_data <- read_csv(paste("./zip/",files[1],sep=""))
  
  if(length(files)>1){
    for(j in 2:length(files)){
      trip_data_j <- read_csv(paste("./zip/",files[j],sep=""))
      trip_data <- rbind(trip_data,trip_data_j)
      rm(trip_data_j)
    }
    file.remove(paste("./zip/",files,sep=""))
  }
  
  
  ###other files
  
  if(length(keys)>1){
    for(i in 1:length(keys)){
      
      save_object(keys[i], file = keys[i], bucket = bucket_id)
      unzip(keys[i],exdir = "./zip")
      file.remove(keys[i])
      files <- list.files("./zip") 
      files <- files[!grepl('MACOSX',files)]
      
      for(j in 1:length(files)){
        trip_data_j <- read_csv(paste("./zip/",files[j],sep=""))
        trip_data <- rbind(trip_data,trip_data_j)
        rm(trip_data_j)
      }
      file.remove(paste("./zip/",files,sep=""))
      
    }
  }
  
  saveRDS(trip_data,"trip_data.RDS")


if(file.exists("weather_data.RDS")){
 weather_data <- readRDS("weather_data.RDS")
}else{
  files <- list.files("./weather/")
  weather_data <- read.csv(paste("./weather/",files[1],sep=""))

  if(length(files)>1){
    for(j in 2:length(files)){
      weather_data_j <- read.csv(paste("./weather/",files[j],sep=""))
      message("read")
      weather_data <- gtools::smartbind(weather_data,weather_data_j)
      rm(weather_data_j)
    }
  }


  

  weather_data_station <- weather_data %>% group_by(STATION,DATE) %>%
  summarise(TAVG=median(TAVG,na.rm = TRUE), #avg temp
            TMIN=median(TMIN,na.rm = TRUE), #min temp
            TMAX=median(TMAX,na.rm = TRUE), #mav temp
            PRCP=median(PRCP,na.rm = TRUE), #precipitation
            SNOW=median(SNOW,na.rm = TRUE), #snowfall
            AWND=median(AWND,na.rm = TRUE), #avg wind speed
            WSF2=median(WSF2,na.rm = TRUE),#fastest 2 min gust
            WDMV=mean(WDMV,na.rm = TRUE), #total wind movement
            WT01=sum(WT01,na.rm = TRUE), #WT01 - Fog, ice fog, or freezing fog (may include heavy fog)
            WT02=sum(WT02,na.rm = TRUE), #WT02 - Heavy fog or heaving freezing fog 
            WT03=sum(WT03,na.rm = TRUE), #WT03 - Thunder
            WT04=sum(WT04,na.rm = TRUE),  #WT04 - Ice pellets, sleet, snow pellets, or small hail
            WT05=sum(WT05,na.rm = TRUE), #WT05 - Hail (may include small hail)
            WT06=sum(WT06,na.rm = TRUE), #WT06 - Glaze or rime
            WT08=sum(WT08,na.rm = TRUE), #WT08 - Smoke or haze
            WT09=sum(as.numeric(WT09),na.rm = TRUE), #WT09 - Blowing or drifting snow
            WT11=sum(WT11,na.rm = TRUE), #WT11 - High or damaging winds
            .groups = 'drop')  %>% 
            mutate(WT01=if_else(is.na(WT01),0,WT01),
                   WT02=if_else(is.na(WT02),0,WT01),
                   WT03=if_else(is.na(WT03),0,WT01),
                   WT04=if_else(is.na(WT04),0,WT01),
                   WT05=if_else(is.na(WT05),0,WT01),
                   WT06=if_else(is.na(WT06),0,WT01),
                   WT08=if_else(is.na(WT08),0,WT01),
                   WT09=if_else(is.na(WT09),0,WT01),
                   WT11=if_else(is.na(WT11),0,WT01),
            ) %>%
            mutate(Date=as_date(parse_date_time(DATE, orders = "Y-m-d"))) %>% select(-DATE)
  
    weather_data <- weather_data %>% group_by(DATE) %>%
    summarise(TAVG=median(TAVG,na.rm = TRUE), #avg temp
            TMIN=median(TMIN,na.rm = TRUE), #min temp
            TMAX=median(TMAX,na.rm = TRUE), #mav temp
            PRCP=median(PRCP,na.rm = TRUE), #precipitation
            SNOW=median(SNOW,na.rm = TRUE), #snowfall
            AWND=median(AWND,na.rm = TRUE), #avg wind speed
            WSF2=median(WSF2,na.rm = TRUE),#fastest 2 min gust
            WDMV=mean(WDMV,na.rm = TRUE), #total wind movement
            WT01=sum(WT01,na.rm = TRUE), #WT01 - Fog, ice fog, or freezing fog (may include heavy fog)
            WT02=sum(WT02,na.rm = TRUE), #WT02 - Heavy fog or heaving freezing fog 
            WT03=sum(WT03,na.rm = TRUE), #WT03 - Thunder
            WT04=sum(WT04,na.rm = TRUE),  #WT04 - Ice pellets, sleet, snow pellets, or small hail
            WT05=sum(WT05,na.rm = TRUE), #WT05 - Hail (may include small hail)
            WT06=sum(WT06,na.rm = TRUE), #WT06 - Glaze or rime
            WT08=sum(WT08,na.rm = TRUE), #WT08 - Smoke or haze
            WT09=sum(as.numeric(WT09),na.rm = TRUE), #WT09 - Blowing or drifting snow
            WT11=sum(WT11,na.rm = TRUE), #WT11 - High or damaging winds
            .groups = 'drop') %>% 
            mutate(WT01=if_else(is.na(WT01),0,WT01),
                   WT02=if_else(is.na(WT02),0,WT01),
                   WT03=if_else(is.na(WT03),0,WT01),
                   WT04=if_else(is.na(WT04),0,WT01),
                   WT05=if_else(is.na(WT05),0,WT01),
                   WT06=if_else(is.na(WT06),0,WT01),
                   WT08=if_else(is.na(WT08),0,WT01),
                   WT09=if_else(is.na(WT09),0,WT01),
                   WT11=if_else(is.na(WT11),0,WT01),
            ) %>%
            mutate(Date=as_date(parse_date_time(DATE, orders = "Y-m-d"))) %>% select(-DATE)
  
  saveRDS(weather_data_station,"weather_data_station.RDS")  
  saveRDS(weather_data,"weather_data.RDS")
}

#No need to rds ..local small file, manually created from google restuls
public_holidays <- read_csv("public_holidays.csv")

public_holidays <- public_holidays %>% mutate(Date=as_date(parse_date_time(Date, orders = "d b Y")))


if(file.exists("bike_stations.RDS")){
  bike_stations <- readRDS("bike_stations.RDS")
}else{

  bike_stations <- rgdal::readOGR("https://opendata.arcgis.com/datasets/a1f7acf65795451d89f0a38565a975b3_5.geojson")
  saveRDS(bike_stations,"bike_stations.RDS")
}

station_daily <- trip_data %>%
  mutate(Date=as_date(`Start date`),
         `Start station number`=as.numeric(`Start station number`)) %>% 
  group_by(Date,`Start station number`,`End station number`) %>%
  summarise(n=n(),.groups = 'drop') %>% ungroup() 


station_demand <- station_daily  %>%  mutate(Station=`Start station number`) %>%
  group_by(Date,Station) %>%
  summarise(n=sum(n),.groups = 'drop') %>% ungroup() %>%
  mutate(season = metR::season(lubridate::month(Date))) %>%
  left_join(seasons, by="season") %>% select(-season) %>%
  left_join(public_holidays,by="Date") %>%
  mutate(weekend=if_else((lubridate::wday(Date,label=TRUE) %in% c("Sat","Sun")),TRUE,FALSE),
         season=season_name,
         public_holiday=if_else(is.na(Event),FALSE,TRUE,missing=FALSE),
         free_day = ifelse(weekend |public_holiday,"Non-working day","Working day")
         ) %>%
  group_by(free_day,season,Station) %>%
  summarise(daily_avg=mean(n),.groups = 'drop') %>% 
  mutate(Type="Start") %>%
  arrange(-daily_avg) %>% ungroup() %>%
  left_join((as.data.frame(bike_stations) %>% 
              mutate(TERMINAL_NUMBER=as.numeric(TERMINAL_NUMBER)) %>%
              select(Station=TERMINAL_NUMBER,LATITUDE,LONGITUDE)),
           by="Station")


workshop<- station_daily %>% filter(`Start station number`==0) %>% 
  group_by(Date) %>%
  summarise(In=sum(n),.groups = 'drop') %>%
  left_join( station_daily %>% filter(`End station number`==0) %>% 
  group_by(Date) %>%
  summarise(Out=sum(n),.groups = 'drop'), by="Date")


stat <- trip_data %>% mutate(year=lubridate::year(`Start date`)) %>% filter(year==2017)


 daily_demand <- trip_data %>% mutate(Date =as_date(`Start date`)) %>% 
  group_by(Date,`Member type`) %>% summarise(n=n(),.groups = 'drop') %>%
  pivot_wider(id_cols = Date,names_from = `Member type`,values_from = n) %>%
   left_join(public_holidays,by="Date") %>%
  mutate(Total=Casual+Member,
         day=lubridate::wday(Date,label=TRUE),
         weekend=if_else((day %in% c("Sat","Sun")),TRUE,FALSE),
         season=metR::season(month(Date)),
         public_holiday=if_else(is.na(Event),FALSE,TRUE,missing=FALSE)
         ) %>%
  left_join(seasons,by="season") %>%
  mutate(season=season_name) %>% select(-season_name) %>%
  left_join(weather_data,by="Date") %>%
  mutate(too_hot=(TMAX>=35 | TAVG>=30),
         too_cold=(TMIN<=-5 | TAVG<=4),
         too_windy=(AWND>30 |WSF2 >=50),
         too_rainy=(PRCP>15 |SNOW >10),
         too_bad = if_else( WT01>4 |
                     WT02>0 |
                     WT03>2 |
                     WT04>4 |
                     WT05>0 |
                     WT06>0 | 
                     WT08>5 |
                     WT09>3 |
                     WT11>0, TRUE,FALSE),
         Good_Weather=ifelse(!too_hot & !too_cold & !too_windy & !too_rainy & !too_bad,'Good','Bad')
  ) %>%
   pivot_longer(cols=c("Casual","Member"),names_to = "type", values_to = "trips") %>%
   select(-Total) 
 
daily_demand$Good_Weather <-  factor(daily_demand$Good_Weather, levels=c('Good','Bad'))
daily_demand$type <-  factor(daily_demand$type, levels=c('Casual','Member'))
daily_demand$season <-  factor(daily_demand$season, levels=c('winter','spring','summer','autumn'))

bikes_per_day<-trip_data %>%
          mutate(Date=as_date(`Start date`)) %>%
  group_by(Date,`Bike number`) %>%
  summarise(n=n(),.groups = 'drop') %>% ungroup() %>%
  select(-n) %>% group_by(Date) %>%
  summarise(bikes=n(),.groups = 'drop') %>% ungroup()


station_capacity <- station_data$data$stations %>% select(Station=short_name,capacity,lon,lat)
rm(station_data)

bikes_per_station_start <- trip_data %>%
            mutate(Date =as_date(`Start date`)) %>%
            group_by(`Start station number`,Date,`Bike number`) %>%
  summarise(n=n(),duration=mean(Duration),.groups = 'drop') %>% ungroup() %>% select(-n) %>%
  group_by(`Start station number`,Date) %>% 
  summarise(bikes=n(),duration=mean(duration),.groups = 'drop') %>% ungroup() %>%
  select(Date,Station=`Start station number`,start_bikes=bikes,duration) 

bikes_per_station_start <- bikes_per_station_start %>% 
                           left_join(station_capacity,by="Station") %>%
                           filter(!is.na(capacity)) %>%
                            mutate(start_cap=round(start_bikes/capacity,2))


stations <- bikes_per_station_start %>% pull(Station) %>% unique(.)


if(exists("station_demand_mod")){
 rm(station_demand_mod) 
}
for(i in 1:length(stations)){
station_demand_mod_i <- weather_data %>%
                      left_join((bikes_per_station_start %>% 
                                filter(Station %in% stations[i])),
                                by="Date") %>%
                      mutate(start_cap=ifelse(is.na(start_cap),0,start_cap),
                             Station=ifelse(is.na(Station),stations[i],Station)
                               ) 

if(exists("station_demand_mod")){
  station_demand_mod <- rbind(station_demand_mod,station_demand_mod_i)
}else{
  station_demand_mod <- station_demand_mod_i
  
}
}

rm(i,stations,station_demand_mod_i)
station_demand_mod_backup <- station_demand_mod

station_filter <- station_demand_mod_backup %>% 
                  filter(!(start_cap==0)) %>%
                  group_by(Station) %>%
                  summarise(n=n(),.groups = 'drop') %>%
                  arrange(n) 

station_demand_mod_remnant <- station_demand_mod_backup %>% 
                              left_join(station_filter,by="Station") %>%
                              mutate(n=ifelse(is.na(n),0,n))%>%
                              filter(n<0)

station_demand_mod <-  station_demand_mod_backup %>%
                       filter(!(Station %in% (station_demand_mod_remnant %>%
                                                pull(Station) %>% unique(.)))) 


resolution <- 0.3

station_demand_mod <- station_demand_mod %>%
                      left_join(station_demand_mod %>%
                                group_by(Station) %>% 
                                summarise(avg_cap=round(mean(start_cap),2), .groups = 'drop'),
                                by="Station") %>%
                      mutate(start_cap_m=start_cap - avg_cap) %>%
                      mutate(start_cap_m=round(start_cap_m/resolution)*resolution)



station_demand_mod <- station_demand_mod %>%
                      left_join(public_holidays,by="Date") %>%
                      mutate(condition1=WT04+WT05+WT06+WT08+WT09+WT11) %>%
                      mutate(day=lubridate::wday(Date,label=TRUE),
                            weekend=if_else((day %in% c("Sat","Sun")),TRUE,FALSE),
                            month=lubridate::month(Date,label=TRUE),
                            season=metR::season(month(Date)),
                            public_holiday=if_else(is.na(Event),FALSE,TRUE,missing=FALSE)
                            ) %>%
                      left_join(seasons,by="season") %>%
                      mutate(season=season_name) %>% select(-season_name) %>%
                      mutate(workday=as.numeric(!public_holiday & !weekend)) 


station_demand_mod1 <- station_demand_mod


station_profile <- station_demand_mod1 %>% group_by(Station) %>%  summarise(min=min(start_cap_m),
                                                              max=max(start_cap_m),
                                                              median=median(start_cap_m),
                                                              zeros=sum(start_cap==0),
                                                              sd=sd(start_cap_m),
                                                              .groups = 'drop') %>% ungroup() 


station_profile1 <- station_profile

cluster_n <-8
sp <- station_profile1 %>% select(-Station)

###The, we run GMM as per the manual
fit <- kmeans(sp, cluster_n, iter.max = 50, nstart = 1)

station_profile1$group <- fit$cluster

station_profile <- station_profile %>%
                   left_join(station_profile1 %>% select(Station,group),by="Station") %>%
                  mutate(group=ifelse(is.na(group),cluster_n+1,group)) %>%
                   select(Station,group)
  




}
trips_stat <- round(nrow(stat)/10^6,2)
bike_fleet <- length(stat %>% pull(`Bike number`) %>% unique(.))
}

```

## Introduction and Purpose
:::::: {.columns}
::: {.column width="60%"}

As the Council-owned bike sharing scheme for the City, *BikeShare* has the mission to provide a reliable, cost-effective bicycle sharing service across the Metropolitan area.  In 2017 alone, *BikeShare* served over 
`r sprintf("%.1f %%", trips_stat)` million trips with a fleet of `r formatC(bike_fleet, format="f", big.mark=",", digits=0)` bicycles.

As of today, bike repairs only occur when a bike is detected broken. Apart from the obvious impact on customer satisfaction when a bike breaks, this model is also inefficient from a workshop’s workload perspective, affecting workload and time to repair.

In order to address, this BikeShare can tap into the existing operational data and use analytics to optimise this problem.


:::
::: {.column width="40%"}

```{r traffic, echo=FALSE,include=FALSE}


fill.breaks <- quantile(station_demand$daily_avg,seq(0,1,0.2)) 
fill.colors <- RColorBrewer::brewer.pal(length(fill.breaks),"YlOrRd")


station_demand$`Daily Avg Traffic` <- cut(station_demand$daily_avg, fill.breaks)

station_demand <- station_demand %>% filter(!is.na(`Daily Avg Traffic`) &
                                            !is.na(LONGITUDE) &
                                              !is.na(LATITUDE))
                                        

bbox <- make_bbox(LONGITUDE, LATITUDE, station_demand, f = 0.005)
m <- get_map(bbox, source = "stamen", zoom = 13)

map <- ggmap(m)  +
  geom_point(data=station_demand,aes(x = LONGITUDE, y = LATITUDE,colour=`Daily Avg Traffic`)) +
  labs(title="Daily trips per Station") +
    theme(axis.title.x=element_blank(),
        axis.text.x=element_blank(),
        axis.ticks.x=element_blank(),
        axis.title.y=element_blank(),
        axis.text.y=element_blank(),
        axis.ticks.y=element_blank()) +
  scale_color_manual(values = fill.colors,
                         breaks =  waiver())  +
  facet_wrap(~season)

rm(m,bbox,fill.colors,fill.breaks)

```

```{r map, dpi = 800, dev.args = list(bg = 'transparent')}
map
```

:::
::::::

## Data analysis of fleet and station  utilisation

:::::: {.columns}
::: {.column width="40%"}
```{r}
fleet_size <- nrow(stat %>% select(`Bike number`) %>% unique(.) )
zeros_stat <- round(nrow(station_profile1 %>% filter(zeros<=20))*100/nrow(station_profile1),0)
zeros_stat2 <-   round(nrow(station_profile1 %>%
                              filter(zeros>=nrow(weather_data)*1/2))*100/nrow(station_profile1),0)
 quarter_stat <-   round(nrow(station_demand_mod_backup %>% group_by(Station) %>% summarise(avg_cap=mean(start_cap),.groups="drop") %>%
  filter(avg_cap<0.25))/nrow(station_demand_mod_backup %>% select(Station)%>% unique(.))*100,0)           
```

However, when further analysis has been conducted, it has become apparent that the previous problem statement is **not** relevant:

* The fleet (`r sprintf("%.0f %%", fleet_size) ` bicycles) exceeds by far its daily demand. 
* Only a fraction (`r zeros_stat` %) of all stations are used daily (or nearly daily). Over a quarter (`r zeros_stat2 ` %) of all stations are half of the time or less. 
* Nearly `r quarter_stat` percent of the stations operate at an average utilisation of less than 25%, while a similar amount operates with an average utilisation of over 75%.


:::
::: {.column width="60%"}
```{r}


dl <- daily_demand %>% group_by(Date) %>%
  summarise(trips=sum(trips),.groups = 'drop') %>%
  ungroup() %>%
  left_join(bikes_per_day,by="Date")

graph1 <- dl %>% ggplot(aes(trips,bikes)) + geom_point(color="lightblue") +
  geom_hline(yintercept=fleet_size, linetype="dashed", color = "red") +
  theme_fivethirtyeight() +
  labs(title="Daily trips and unique bicycle use - 2017 and 2018",
       x="Daily Trips",
       y="Unique number of bikes used each day") +
  theme(legend.position = "bottom",
        plot.title = element_text(size=12),
        axis.title.x = element_text(size = 10),
        axis.text.x = element_text(angle = 0, hjust = 1,size = 8),
        axis.title.y = element_text(size = 10),
        axis.text.y = element_text(size = 8),
        strip.text.x = element_text(size = 8),
        strip.text.y = element_text(size = 8, angle = 90),
        legend.title=element_text(size=8),
        legend.text=element_text(size=8)) +
  scale_y_continuous(labels=function(x) format(x, big.mark = ",", scientific = FALSE))+
 scale_x_continuous(labels=function(x) format(x, big.mark = ",", scientific = FALSE))

graph2 <- station_profile1 %>% ggplot(aes(zeros))+
   geom_histogram( colour="navy", fill="lightblue",binwidth = 200) +
  theme_fivethirtyeight() +
  labs(title="Distribution of Stations by number of days with zero used",
       x="Number of day with zero bikes used",
       y="Unique bike stations") +
  theme(legend.position = "bottom",
        plot.title = element_text(size=12),
        axis.title.x = element_text(size = 10),
        axis.text.x = element_text(angle = 0, hjust = 1,size = 8),
        axis.title.y = element_text(size = 10),
        axis.text.y = element_text(size = 8),
        strip.text.x = element_text(size = 8),
        strip.text.y = element_text(size = 8, angle = 90),
        legend.title=element_text(size=8),
        legend.text=element_text(size=8)) +
  scale_y_continuous(labels=function(x) format(x, big.mark = ",", scientific = FALSE))+
 scale_x_continuous(labels=function(x) format(x, big.mark = ",", scientific = FALSE))

graph1 / graph2
```

:::
::::::

## Further Analysis : There is a geographic station gap.
:::::: {.columns}
::: {.column width="60%"}

In order to perform further analysis, bike stations were clustered using  their key statistical measures. After reviewing the results, they are were aggregated into three distinct usage groups:

* **High use**. These stations are frequently use and during summer they may carry several times their capacity. They exhibit the highest seasonality and variation. They are mostly located in the inner city and dense suburban pockets, and they are close to each other.
* **Medium use**. Stations with lower average usage compared with the high use group. They have a slightly lower seasonality, They are located in inner city districts (sometime opposite the street from high use locations) or along peri-urban commuting corridors and dense near-city suburban centres.
* **Low use**. These stations have very low usages, and many of them go unused for days. They are located in the city but also  peripheral suburbs. The low volume of data makes it difficult to forecast demand.

It is interesting to note that stations of the three categories share the inner city. This makes any demand prediction challenging, due to utilisation influence by other variables difficult to quantify, such as proximity to key landmark of public transport exit.


:::
::: {.column width="40%"}

```{r echo=FALSE, warning=FALSE,message=FALSE}
map_data <- station_capacity %>% 
  left_join(station_profile1, by="Station") %>%
  mutate(business_group=if_else(group==7,"high use",
                                ifelse(group==4,"medium use","low use"),
                                missing="low use"))

bbox <- make_bbox(LONGITUDE, LATITUDE, station_demand, f = 0.005)
m <- get_map(bbox, source = "stamen", zoom = 13)

map <- ggmap(m)  +
  geom_point(data=map_data,aes(x = lon, y = lat,colour=business_group)) +
  labs(title="Station Groups") +
    theme(axis.title.x=element_blank(),
        axis.text.x=element_blank(),
        axis.ticks.x=element_blank(),
        axis.title.y=element_blank(),
        axis.text.y=element_blank(),
        axis.ticks.y=element_blank()) +
      scale_color_discrete() + facet_wrap(.~ business_group)

map
```
:::
::::::

## Results and recommendations

Based on the previous analysis, we can confirm that maintenance planning and demand utilisation are not priority items to address, given the existing fleet size and location in each station. 

Nevertheless, it has become clear that the points to tackle are the under-utilisation of the fleet and disparate station usage. Therefore, we present the below recommendations:

* Address the large number of spares by reducing the fleet or leveraging this asset in the below recommendations.
* Conduct experiments to redirect demand in inner city to medium and low use stations.
* After redirection attempts, optimise location and size or inner city stations. Optimise the peripheral network.
* Use excess capacity to deploy "pop up" stations on high usage events and locations (e.g. offer them as a public transport alternative during massive events, bring bikes to parks in spring).
* Explore innovate bike use models/service offerings, such as:
** Multi-day (weekend) trips for registered users.
** Allow suburban commuters to take bike home overnight - taking advantage of excess bikes.
** Allow bikes to be parked on regular bike parking spots - to reduce walking distances.
** Explore park&ride scheme with car parks in the inner city fringe.
** Explore ride & train scheme for suburbs.


## Appendix

The below slides were used in previous stages and are presented here for reference purposes

## Fleet maintenance volumes and windows need to be carefully chosen
:::::: {.columns}
::: {.column width="40%"}

In order to **<span style='color: brown;'>provide a good service</span>** for the users of the bike sharing scheme, **<span style='color: brown;'>unavailability needs to be avoided</span>**. Unavailability is caused by lack of maintenance or lack of supply. Thus, maintenance windows need to be selected **<span style='color: brown;'>smartly</span>**.

- Shared bikes are **<span style='color: brown;'>used every day of the week</span>**. Registered users provide a constant demand baseline.
- There is **<span style='color: brown;'>some seasonality</span>** in the demand: registered users brave through the winter keeping the numbers high.
- At first look, it seems that **<span style='color: brown;'>weather conditions are the biggest deterrent</span>** for bike usage. If it is too cold, or too hot or too windy casual users won't ride; registered users' numbers will also drop.
- Bikes will **<span style='color: brown;'>keep breaking</span>** - even if they are well built! Thus, **<span style='color: brown;'>every opportunity</span>** to undertake maintenance **<span style='color: brown;'>needs to be taken</span>**.
:::
::: {.column width="60%"}

```{r problem_statement_chart,dpi = 800, dev.args = list(bg = 'transparent'), echo = FALSE}
text_value <-   "**<span style='color: brown;'>Challenge</span>**: Using data analytics, we can leverage the **<span style='color: brown;'>weather forecast </span>** to optimise the fleet **<span style='color: brown;'>maintenace without compromising availibilty SLAs</span>**."





g <-daily_demand %>%
  ggplot(aes(x=day,y=trips,color=Good_Weather, group=1)) +
  geom_jitter(shape=16, position=position_jitter(0.2)) +
  facet_grid(type ~ season, space='free_y') +
 theme_fivethirtyeight() + 
  scale_shape_tableau() +
  scale_colour_tableau('Classic Color Blind') +
  scale_y_continuous(labels = comma) +
  theme(legend.position = "bottom",
        plot.title = element_text(size=14),
        axis.title.x = element_text(size = 10),
        axis.text.x = element_text(angle = 90, hjust = 1,size = 8),
        axis.title.y = element_text(size = 10),
        axis.text.y = element_text(size = 8),
        strip.text.x = element_text(size = 8),
        strip.text.y = element_text(size = 8, angle = 90),
        legend.title=element_text(size=8),
        legend.text=element_text(size=8)) +
  labs(title="Bike Demand per Season and User Type",
       x="Day of the Week",
       y="Bikes Rented") +
  guides(col = guide_legend(ncol = 2),fill = guide_legend(title = "Weather", title.position = "left"))


t<-  textbox_grob(
  text_value,
  gp = gpar(fontsize = 12),
  box_gp = gpar(col = "grey80", fill = NA,lineend="round",lwd=3),
  r = unit(10, "pt"),
  padding = unit(c(10, 10, 10, 10), "pt"),
  width = unit(1, "npc"),
#  height  = unit(1, "npc"),
  margin = unit(c(10, 10,10, 10), "pt")
)

layout <- "
AAAA
AAAA
AAAA
AAAA
AAAA
####
CCCC
CCCC
"
g + t + plot_layout(design = layout) 
rm(g,t,layout,text_value)

```
:::
::::::


## By addressing uneven bike utilisation, it is possible to extend the life of the fleet
:::::: {.columns}
::: {.column width="40%"}

Some bicycles in the fleet are heavily used almost every single days, while others are seldomly ridden. As a result:

 - **<span style='color: brown;'>Heavily used bikes will be at higher risk of failure</span>** and accelerated asset ageing.
 - **<span style='color: brown;'>Unused bikes will be at risk of neglect</span>** - e.g. left with flat tyres in an low traffic station.
 
By obtaining a detailed log of all bike travel, it will be possible to identify which bikes are being used the most and calculate their risk of failing. This will allow to proactively book them for service.  

When those bike are taken out for service, they can be replaced with lower used bikes , evening asset utilisation.


:::
::: {.column width="60%"}

```{r bike usage, dpi = 800, dev.args = list(bg = 'transparent')}
t<- trip_data %>% mutate(Date=as_date(`Start date`)) %>%
  group_by(Date,`Bike number`) %>%
  summarise(trips=n(),.groups = 'drop') %>% ungroup()
  
t<- t %>% mutate(Month=format(as.Date(Date), "%Y-%m")) %>%
  filter(Month %in% c("2017-05","2017-06","2017-07","2017-08"))

selection <- t %>% group_by(`Bike number`) %>% summarise(n=n(),.groups = 'drop') %>% ungroup() %>% arrange(-n)


selection2 <- rbind(rbind(selection[1:7,],
                   selection[(nrow(selection)-6):nrow(selection),]),
                   selection[(round(nrow(selection)/2,0)-3):(round(nrow(selection)/2,0)+3),])
                   

bike_use <- t %>% filter(`Bike number` %in% selection2$`Bike number`) %>%
  ggplot(aes(Date,`Bike number` , fill= trips)) + 
  geom_tile()+
  scale_fill_gradient(low="lightblue", high="navy") +
   theme_fivethirtyeight() +
  labs(title="Daily trips - selected bikes from May to Aug 2017",
       x="Date",
       y="Bike Number") +
  theme(legend.position = "bottom",
        plot.title = element_text(size=12),
        axis.title.x = element_text(size = 10),
        axis.text.x = element_text(angle = 90, hjust = 1,size = 8),
        axis.title.y = element_text(size = 10),
        axis.text.y = element_text(size = 8),
        strip.text.x = element_text(size = 8),
        strip.text.y = element_text(size = 8, angle = 90),
        legend.title=element_text(size=8),
        legend.text=element_text(size=8)) 


rm(t,selection,selection2)

bike_use
```

:::
::::::

## Data Collection to Optimise Maintenance workload and bike availability
:::::: {.columns}
::: {.column width="40%"}

To achieve the objectives, the below data needs to be collected:

- [Detailed trip data, itemised per each trip taken by each bicycle](https://www.capitalbikeshare.com/system-data)
- [Weather Observations/Forecast for each day of the year](https://www.ncdc.noaa.gov/cdo-web/datasets)
- Bike failure rates.
- Workshop's repair turnaround stats.

:::
::: {.column width="40%"}

Based on the data, the below metrics will be generated:

```{r table}

table_data <- data.frame(Metric=c("Daily Service Capacity",
                             "Need to Service Factor"),
                Description=c("Number of bikes that can be maintain any day, based on workshop capacity and predicted demand",
                              "Indicator whether a bike needs service ahead of probable failure"),
                     Source=c("Workshop Turnaround stats, Summary of daily trip data, weather",
                              "Detailed trip data")
                     )

table <- table_data %>% flextable() %>%
  width(j=1,width=1.5) %>%
  width(j=2:3,width=3) %>%
  bold(i = NULL, j = 1, bold = TRUE, part = "body") %>%
  bold(i = NULL, j = NULL, bold = TRUE, part = "header") %>%
  align(i = NULL, j = NULL, align = "left", part = "all") %>%
  border(
  i = NULL,
  j = NULL,
  border = NULL,
  border.top = NULL,
  border.bottom = fp_border(color = "grey"),
  border.left = NULL,
  border.right = NULL,
  part = "body"
)

knitr::knit_print(table)

```

These metrics (combined with spontaneous failure rates) should serve as input for a proactive service plan.

:::
::::::

## Data Sources

- Bike scheme data: https://www.capitalbikeshare.com/system-data
- Weather data: https://www.ncdc.noaa.gov/cdo-web/search;jsessionid=FC6D9E3A1F49486E29E7130142A60038