south ural state university, Russian federation
# Imports
library(fpp2)
## Warning: package 'fpp2' was built under R version 4.0.3
## Registered S3 method overwritten by 'quantmod':
## method from
## as.zoo.data.frame zoo
## -- Attaching packages ------------------------------------------------------------------------------ fpp2 2.4 --
## v ggplot2 3.3.2 v fma 2.4
## v forecast 8.13 v expsmooth 2.3
## Warning: package 'ggplot2' was built under R version 4.0.3
## Warning: package 'forecast' was built under R version 4.0.3
##
library(forecast)
library(ggplot2)
dataset_path<-"F:/Phd/Private Project/train_1-slice0.csv" #path of your time series data in your pc
y_lab<- "Number Vistors Apple page in wikipedia" # input name of Time series data
Actual_date_interval <- c("2015/07/01","2016/12/31") # Write your actual interval date of your time series
Forecast_date_interval <- c("2017/01/01","2017/01/7") # Write your Forecasting interval of the date your time series
validation_data_days <-7 # how many rows from actual data that you want to test the TBATS model and calculate MAPE for that perioud
frequency<-"days" # Type of your data ("years","months","weeks","days") that should you choose type of your data
# Data Preparation
original_data<-read.csv(dataset_path)
# Apple Page : original_data$Apple_II_zh.wikipedia.org_all.access_spider
# Facebook Page : original_data$Facebook_zh.wikipedia.org_all.access_spider
# Android Page : original_data$Android_zh.wikipedia.org_all.access_spider
# Energy Page : original_data$Energy_zh.wikipedia.org_all.access_spider
input_data<-cumsum(original_data$Apple_II_zh.wikipedia.org_all.access_spider) # convert data to cumulative
rows <- NROW(input_data)
training_data<-input_data[1:(rows-validation_data_days)]
testing_data<-input_data[(rows-validation_data_days+1):rows]
AD<-fulldate<-seq(as.Date(Actual_date_interval[1]),as.Date(Actual_date_interval[2]), frequency) #input range for actual date
FD<-seq(as.Date(Forecast_date_interval[1]),as.Date(Forecast_date_interval[2]), frequency) #input range forecasting date
N_forecasting_days<-nrow(data.frame(FD))
validation_dates<-tail(AD,validation_data_days)
validation_data_by_name<-weekdays(validation_dates)
forecasting_data_by_name<-weekdays(FD)
# Data Modeling
data_series<-ts(training_data)
model_holt<-holt(data_series,h=N_forecasting_days+validation_data_days,lambda = "auto")
accuracy(holt(data_series,h=N_forecasting_days+validation_data_days,lambda = "auto")) # accuracy on training data
## ME RMSE MAE MPE MAPE MASE
## Training set -0.1331319 5.00807 3.324313 0.1261135 0.6548812 0.3745121
## ACF1
## Training set 0.02402595
# Print Model Parameters
plot(model_holt,main =paste(y_lab),xlab = paste ("Time in ", frequency , sep=" "),ylab = y_lab)
# Testing Data Evaluation
forecasting_holt <- predict(model_holt, h=N_forecasting_days+validation_data_days,lambda = "auto")
validation_forecast<-head(forecasting_holt$mean,validation_data_days)
MAPE_Per_Day<-round( abs(((testing_data-validation_forecast)/testing_data)*100) ,3)
paste ("MAPE % For ",validation_data_days,frequency,"by using holt Model for ==> ",y_lab, sep=" ")
## [1] "MAPE % For 7 days by using holt Model for ==> Number Vistors Apple page in wikipedia"
MAPE_Mean_All<-paste(round(mean(MAPE_Per_Day),3),"% MAPE ",validation_data_days,frequency,y_lab,sep=" ")
MAPE_holt<-paste(round(MAPE_Per_Day,3),"%")
MAPE_holt_Model<-paste(MAPE_Per_Day ,"%")
paste (" MAPE that's Error of Forecasting for ",validation_data_days," days in holt Model for ==> ",y_lab, sep=" ")
## [1] " MAPE that's Error of Forecasting for 7 days in holt Model for ==> Number Vistors Apple page in wikipedia"
paste(MAPE_Mean_All,"%")
## [1] "0.363 % MAPE 7 days Number Vistors Apple page in wikipedia %"
paste ("MAPE that's Error of Forecasting day by day for ",validation_data_days," days in holt Model for ==> ",y_lab, sep=" ")
## [1] "MAPE that's Error of Forecasting day by day for 7 days in holt Model for ==> Number Vistors Apple page in wikipedia"
data.frame(date_holt=validation_dates,validation_data_by_name,actual_data=testing_data,forecasting_holt=validation_forecast,MAPE_holt_Model)
## date_holt validation_data_by_name actual_data forecasting_holt
## 1 2016-12-25 воскресенье 4826 4823.601
## 2 2016-12-26 понедельник 4834 4832.202
## 3 2016-12-27 вторник 4858 4840.803
## 4 2016-12-28 среда 4868 4849.405
## 5 2016-12-29 четверг 4883 4858.007
## 6 2016-12-30 пятница 4895 4866.610
## 7 2016-12-31 суббота 4906 4875.214
## MAPE_holt_Model
## 1 0.05 %
## 2 0.037 %
## 3 0.354 %
## 4 0.382 %
## 5 0.512 %
## 6 0.58 %
## 7 0.628 %
data.frame(FD,forecating_date=forecasting_data_by_name,forecasting_by_holt=tail(forecasting_holt$mean,N_forecasting_days))
## FD forecating_date forecasting_by_holt
## 1 2017-01-01 воскресенье 4883.818
## 2 2017-01-02 понедельник 4892.422
## 3 2017-01-03 вторник 4901.027
## 4 2017-01-04 среда 4909.633
## 5 2017-01-05 четверг 4918.239
## 6 2017-01-06 пятница 4926.845
## 7 2017-01-07 суббота 4935.452
plot(forecasting_holt,xlab = paste ("Time in ", frequency , sep=" "),ylab = y_lab)
x1_test <- ts(testing_data, start =(rows-validation_data_days+1) )
lines(x1_test, col='red',lwd=2)
graph<-autoplot(forecasting_holt,xlab = paste ("Time in ", frequency , sep=" "),ylab = y_lab)
graph
