Прогнозирование временных рядов с использованием R - PullRequest
Новая
       7

Прогнозирование временных рядов с использованием R

0 голосов
/ 03 мая 2018

У меня проблема с прогнозированием этих нестационарных данных (https://drive.google.com/file/d/14o5hHe8zxR0onRWq0mZNcYqI101O0Dkw/view?usp=sharing) с использованием Auto Arima. Пожалуйста, просмотрите мой код. 

# Read Data
r = read.csv('../Amazon/Amazon1.csv', header = TRUE, stringsAsFactors = FALSE)
# Time Series construction
ts = ts(t(r[,1:25]), frequency = 12, start = c(2016,01) )
# plotting Time series

ts.plot(ts[,1:2],type = 'b', xlab = 'Monthly Cycle', ylab = 'Number of Sales', main = "(TIME SERIES) Amazon Sales Cycle of multiple products for 24 months",col=c(rep("black",1),rep("red",2)))
legend("topleft",cex=.65,legend = ts[1,1:2], col = 1:ncol(ts), lty = 1)

set1 = ts[2:20,1]
#set2 = ts[15:20,1]
set1 = as.numeric(set1)
#set2 = as.numeric(set2)

# Building Forecasting models
mf = meanf(set1,h=4,level=c(90,95),fan=FALSE,lambda=NULL) 
plot(mf)

mn = naive(set1,h=4,level=c(90,95),fan=FALSE,lambda=NULL) 
plot(mn)

md = rwf(set1,h=4,drift=T,level=c(90,95),fan=FALSE,lambda=NULL) 
plot(md)
# Checking Accuracy
accuracy(mf)
accuracy(mn)
accuracy(md)
# Identifying Stationarity/Non-Stationarity(unit Root testing)
adf = adf.test(set1)
adf

kpss = kpss.test(set1)
kpss
ndiffs(set1)
diff_data = diff(set1)

adf.test(diff_data) # Rerunning unit test on differenced data
# Identifying Seasonality/Trend
Stl = stl(set1,s.window='periodic')

# ARIMA modelling

ar_set1 = forecast::auto.arima(diff(diff_data), approximation=FALSE,trace=FALSE, stationary = TRUE)
forecast(ar_set1, h = 5)
ar_set2 = forecast::ets(diff(diff_data))
ar_set3 = forecast::nnetar(diff(diff_data), approximation=FALSE,trace=FALSE, stationary = TRUE)
# Prediction
predict(ar_set1, n.ahead = 5,se.fit = TRUE)
predict(ar_set2, n.ahead = 5,se.fit = TRUE)
predict(ar_set3, n.ahead = 5,se.fit = TRUE)
plot(forecast(ar_set1,h=6))
points(1:length(diff(diff_data)),fitted(ar_set1),type="l",col="green")

plot(forecast(ar_set2,h=6))
points(1:length(diff(diff_data)),fitted(ar_set2),type="l",col="green")

plot(forecast(ar_set3,h=6))
points(1:length(diff(diff_data)),fitted(ar_set3),type="l",col="green")



accuracy(ar_set1, test = diff_data)
accuracy(ar_set2, test = diff_data)
accuracy(ar_set3, test = diff_data)

Я не могу получить прогнозные значения одним из этих трех методов. Куда я иду не так?

...