modeltime r package for time series forecasts - google analytics data

modeltime r package for time series forecasts.r

# R TIPS ----
# TIP 040 | Introduction to Modeltime: In Under 10-Minutes ----
#By Matt Dancho, modeltime package creator
#R Tips newsletter by Matt Dancho: https://mailchi.mp/business-science/r-tips-newsletter

# LIBRARIES ----
#installing packages from git 
#install.packages("devtools")
library(devtools)
library(usethis)

#install_github("tidymodels/tidymodels")
#install.packages("modeltime")
#install.packages("tiymodels")
#install_version("modeltime")
library(modeltime)
library(tidymodels)
library(tidyverse)
library(timetk)
library(lubridate)

# DATA ----
#install.packages("googleAnalyticsR")
library(googleAnalyticsR)
#install.packages("ellipsis")
ga_auth(json_file="C:\\folder-where-you-place-your-json-file\\service-key.json")
#the Google Analytics view id
viewID <- 12345678

data <- google_analytics(
    viewID,
    date_range=c("2019-01-01","2021-07-05"),
    dimensions=c("date"),
    metrics=c("sessions"),
    anti_sample=TRUE
)
#backup data to xlsx
write_xlsx(data,path="C:\\Users\\folder-where-you-want-to-write-file-to\\data.xlsx",
           col_names = TRUE,
           format_headers = TRUE,
           use_zip64 = FALSE)

summary(data)
View(data)

#plot time series data
ga_data %>% plot_time_series(date,sessions)

#split data into training /test
splits <- time_series_split(
    ga_data,
    assess     = "6 months",
    cumulative = TRUE
)

#plot training and test ts
splits %>%
    tk_time_series_cv_plan() %>%
    plot_time_series_cv_plan(date, sessions)

# FORECAST ----

# * AUTO ARIMA ----
library(prophet)
?prophet()
library(forecast)
model_arima <- arima_reg() %>%
    set_engine("auto_arima") %>%
    fit(sessions ~ date, training(splits))

model_arima

# * Prophet ----
model_prophet <- prophet_reg(
    mode="regression",
    seasonality_weekly = "auto",
    seasonality_yearly = "auto",
    seasonality_daily = "auto"
) %>%
    set_engine("prophet") %>%
    fit(sessions ~ date, training(splits))

model_prophet

#AUTO ETS Model
model_ets <- exp_smoothing()%>%
    set_engine("ets") %>%
    fit(sessions ~ date,training(splits))

model_ets


#Snaive model
model_snaive <- naive_reg()%>%
    set_engine("snaive")%>%
    fit(
    sessions~date,training(splits))
model_snaive

# MODELTIME COMPARE ----

# * Modeltime Table ----
model_tbl <- modeltime_table(
    model_arima,
    model_prophet,
    model_ets,
    model_snaive
)

# * Calibrate ----
calib_tbl <- model_tbl %>%
    modeltime_calibrate(testing(splits))

# * Accuracy ----
calib_tbl %>% modeltime_accuracy()

# * Test Set Visualization ----
calib_tbl %>%
    modeltime_forecast(
        new_data    = testing(splits),
        actual_data = data,
        conf_interval = 0
    ) %>%
    plot_modeltime_forecast()

# * Forecast Future ----
future_forecast_tbl <- calib_tbl %>%
    modeltime_refit(data) %>%
    modeltime_forecast(
        h           = "3 months",
        actual_data = data,
        conf_interval = 0
    )
#plot forecast
future_forecast_tbl %>%
    plot_modeltime_forecast()