Tims serie analysis

# Load packages
if (!require("pacman")) install.packages("pacman") # allow easy package installation

# load required packages
pacman::p_load(tidyverse, # for data wrangling, visualization, etc 
               janitor,   # for data cleaning
               forecast)      # for time serie analysis

theme_set(theme_minimal()) #set the theme for all the graphs

Dataset

df <- read_csv("Event_Segmentation.csv", skip = 6) # read the csv and skipe the first n rows

Check the dataset

head(df)

Transpose the dataset

df <- df %>%
  pivot_longer(-Event, names_to = "date", values_to = "n_type") %>% 
  mutate(date = lubridate::ymd(date) ) %>% 
  janitor::clean_names() %>% 
  tibble::rowid_to_column( "id") # a column with unique id

Check the dataset

head(df)

Exploratory Data Analysis (EDA)

df %>%
  ggplot(aes(x = date, 
             y = n_type, 
             color = event)) +
  geom_line() +
  labs(    title = "EDA",
           y = "n",
           x = "Date",
           color = "Event"
  )

df %>%
  ggplot(aes(x = date, 
             y = n_type, 
             color = event)) +
  geom_line() +
  labs(    title = "EDA with regression line and smoothing each 7 days",
           y = "n",
           x = "Date",
           color = "Event"
  ) + 
  geom_smooth(method = "loess", se = TRUE, span = 0.7)

df %>%
  ggplot(aes(x = date, 
             y = n_type, 
             color = event)) +
  geom_line() +
  labs(    title = "EDA",
           y = "n",
           x = "Date",
           color = "Event"
  ) + 
  facet_grid(.~event)

df %>%
  ggplot(aes(x = date, 
             y = n_type, 
             color = event)) +
  geom_line(alpha = .5) +
  labs(    title = "EDA with regression line, smoothing each 7 days",
           y = "n",
           x = "Date",
           color = "Event"
  ) + 
  facet_grid(.~event) + 
  geom_smooth(method = "loess", se = TRUE, span = 0.7)

Grouped by week

df %>% 
  group_by(week = lubridate::week(date)) %>%
  ggplot(aes(x = as.factor(week), y = n_type, color = event)) + 
  geom_boxplot() + 
  facet_grid(.~event)

df %>% 
  group_by(week = lubridate::week(date), event) %>%
  summarise(mean_n = mean(n_type)) %>% 
  pivot_wider(event)

Time serie analysis

Compare each monthly week

pacman::p_load(tibble, 
               feasts, 
               fable)

Convert df to tsible format