1 주요용어

2 시차상관(선행, 후행)

3 plotForecastErrors

4 지수평활법: Holt-Winters filtering

5 이상점 탐지

6 ARIMA

7 AR

8 평활

9 기타

10 VAR

11 auto.arima

12 참고자료

시계열자료는 일반적으로 독립이 아니며, 시간순서에 따라 분석을 해야 한다. 시계열은 4개의 성분으로 분해되는데, 장기적 변화인 추세(Trend: T), 계절변동(Seasonal fluctuation: S), 비계절적 변동을 의미하는 순환(Cyclic: C), 그리고 불규칙적 오차(Random error: R)가 그 성분들이 된다. --네이버 교육용어사전

• trend
  
• seasonal fluctuation
  
• cycles
  
• white noise
  
  

> tmp
   curr_val prev_val
5   3949515  4316313
6   3835690  4149917
7   3975137  4165120
8   3993659  4050347
9   3942373  3949515
10  3825807  3835690
11  3920338  3975137
12  3586568  3993659
13  3633877  3942373
14  3444306  3825807
15  3903889  3920338
16  3529384  3586568
17  3531147  3633877
18  3518490  3444306
19  3481886  3903889
20  3298512  3529384
21  3400325  3531147
22  3641914  3518490
23  3466527  3481886
24  6028886  3298512
25  6543880  3400325
26  5516709  3641914
27  5100715  3466527
28  5259752  6028886
29  5370662  6543880
30  5009019  5516709
31  4778911  5100715
32  4836214  5259752
33  4662136  5370662

tmp <- na.omit(tmp)
ccf(tmp$curr_val, tmp$prev_val)

첫 번째 매개변수를 기준으로 얼마나 선행/후행 하는지 알 수 있다. 가장 볼록하게 나온 부분이 x축에 -4 이니까 tmp$curr_val가 대략 4일정도 빠르다.

첫 번째 매개변수를 기준으로 얼마나 선행하는지 알 수 있다.
lag.plot2(itall.R) 함수를 이용해서 차트를 그려보자.

source("http://databaser.net/moniwiki/pds/TimeSeries/itall.R")
lag.plot2(tmp$curr_val , tmp$prev_val, 11)

표준화해서 라인 차트를 그려보자.

plot(scale(tmp$curr_val), type="l", col="red")
lines(scale(tmp$prev_val), col="blue")

최대/최소 교차상관 및 시차 (출처)

max_ccf <- function(...)
{
    d <- ccf(...)
    cor <- d$acf[,,1]
    lag <- d$lag[,,1]
    dd <- data.frame(cor, lag)
    max_val <- dd[which.max(dd$cor),]
    min_val <- dd[which.min(dd$cor),]
    result <- c()
      if(abs(max_val[1]) >= abs(min_val[1])){
          result <- max_val
      } else {
          result <- min_val
      }
    return(result)
}

max_ccf <- function(...)
{
  d <- ccf(...)
  cor <- d$acf[,,1]
  lag <- d$lag[,,1]
  df <- data.frame(cor, lag)
  max_val <- df[which.max(df$cor),]
  return(max_val)
}

min_ccf <- function(...)
{
  d <- ccf(...)
  cor <- d$acf[,,1]
  lag <- d$lag[,,1]
  df <- data.frame(cor, lag)
  min_ccf <- df[which.min(df$cor),]
  return(min_ccf)
}

예를 들어, 다음과 같이 결과가 나왔다면..

> max_ccf(sales, val)
         cor lag
12 0.8932095  -1

• 최대 상관 계수는 0.89
  
• 시간축 상에서 sales가 val보다 -1일 뒤져 있음을 뜻함.
  
  

plotForecastErrors <- function(forecasterrors)
{
    # make a histogram of the forecast errors:
    mybinsize <- IQR(forecasterrors)/4
    mysd   <- sd(forecasterrors)
    mymin  <- min(forecasterrors) - mysd*5
    mymax  <- max(forecasterrors) + mysd*3
    # generate normally distributed data with mean 0 and standard deviation mysd
    mynorm <- rnorm(10000, mean=0, sd=mysd)
    mymin2 <- min(mynorm)
    mymax2 <- max(mynorm)
    if (mymin2 < mymin) { mymin <- mymin2 }
    if (mymax2 > mymax) { mymax <- mymax2 }
    # make a red histogram of the forecast errors, with the normally distributed data overlaid:
    mybins <- seq(mymin, mymax, mybinsize)
    hist(forecasterrors, col="red", freq=FALSE, breaks=mybins)
    # freq=FALSE ensures the area under the histogram = 1
    # generate normally distributed data with mean 0 and standard deviation mysd
    myhist <- hist(mynorm, plot=FALSE, breaks=mybins)
    # plot the normal curve as a blue line on top of the histogram of forecast errors:
    points(myhist$mids, myhist$density, type="l", col="blue", lwd=2)
}
p10 <- forecast(m, 10)
plotForecastErrors(p10 $residuals)

계절적요인과 불규칙요인을 제거

m1<- HoltWinters(ts, gamma=FALSE)
plot(m1)

plot(forecast.HoltWinters(m1, h=10))

library("TSA")
detectAO(m) #가법적 이상점(AO;additive outlier)을 탐지 - T번째 관측값 수준 관점에의 이상치
detectIO(m) #혁신적 이상점(IO;innovative outlier)을 탐지 - 모든값 관점의 이상치

library("zoo")
ts <- zoo(long_term$dau, as.Date(long_term$dt))
plot(ts)
fit <- auto.arima(ts)
plot(forecast(fit, h=20))

arima(ts, order=c(2,1,2))
m <- arima(ts, order=c(2,1,2), fixed=c(0,NA,0,NA))
confint(m)
tsdiag(m)
predict(m, n.ahead=10)

library("lmtest") 
coeftest(fit) # ARIMA p-value test

월단위?

t <- ts(training$amt, frequency=30)
fit <- auto.arima(t,D=1)
pred <- forecast(fit,h=30)
plot(pred)
grid()

• http://www.r-bloggers.com/simple-and-advanced-time-series-with-oracle-r-enterprise/?utm_source=feedburner&utm_medium=email&utm_campaign=Feed%3A+RBloggers+%28R+bloggers%29&utm_content=Yahoo%21+Mail
  
• http://stat.epfl.ch/files/content/sites/stat/files/shared/Applied%20Statistics/notes10.pdf
  
• http://stats.stackexchange.com/questions/41070/how-to-setup-xreg-argument-in-auto-arima-in-r --> xreg
  
• http://robjhyndman.com/
  
• http://stackoverflow.com/questions/20210504/r-times-series-arima-model-forecasting-daily-data
  
• https://www.otexts.org/fpp/8/9
  
• http://stackoverflow.com/questions/10302261/forecasting-time-series-data
  

library("MASS")
ar.result <- ar(ts)
predict(ar.result, n.ahead=9)

library("forecast")
data(gold)
g <- data.frame(g = na.omit(coredata(gold)))$g
y_width <- c(min(g) - min(g) * 0.05, max(g) - max(g) * 0.05) 

plot(g, type="l", lty="dotted", ylim=y_width)
par(new=T)
plot(smooth(g, twiceit=T), type="l", col="blue", ylim=y_width, ylab="")

이동중간값 함수 : runmde()

--그림 출처: http://simplystatistics.org/2014/02/13/loess-explained-in-a-gif/

7일

ts_df <- ts(mydata$y, freq=365.25/7, start=decimal_date(ymd(min(mydata$date_key))))

• http://christophj.github.io/replicating/r/vector-autoregression-var-in-r/
  
• varModels.pdf
  
• 벡터자기회귀_VAR_모형의_이해PDF.pdf
  
  

library(lubridate)
library("forecast")
ts_df <- ts(mydata$y, freq=7)
#ts_df <- ts(mydata$y, freq=365.25/7, start=decimal_date(ymd(min(mydata$dt)))) 
d.arima <- auto.arima(ts_df)
d.forecast <- data.frame(forecast(d.arima, level = c(99), h = 1))

• https://otexts.com/fppkr/
  
• http://www.itc.nl/~rossiter/teach/R/R_ts.pdf
  
• http://a-little-book-of-r-for-time-series.readthedocs.org/en/latest/src/timeseries.html
  
• http://maths-people.anu.edu.au/~johnm/courses/r/ASC2008/pdf/Rtimeseries-ohp.pdf
  
• Time Series Analysis and Its Applications: With R Examples
  
• https://onlinecourses.science.psu.edu/stat510/?q=node/41
  
• itall.R
  
• Time Series Analysis with R - Part I
  
• Quick-R: Time Series