Master R 12 - Analyzing Time-series

Creating time-series objects

library(hflights)

library(data.table)

dt <- data.table(hflights)

dt[, date := ISOdate(Year, Month, DayofMonth)]

daily <- dt[, list(

     N         = .N,

     Delays    = sum(ArrDelay, na.rm = TRUE),

     Cancelled = sum(Cancelled),

     Distance  = mean(Distance)

 ), by = date]

str(daily)

Visualizing time-series

plot(ts(daily))

setorder(daily, date)

plot(ts(daily))

plot(ts(daily$N, start = 2011, frequency = 365), main = 'Number of flights from Houston in 2011')

Seasonal decomposition

plot(decompose(ts(daily$N, frequency = 7)))

setNames(decompose(ts(daily$N, frequency = 7))$figure, weekdays(daily$date[1:7]))

decompose(ts(daily$N, frequency = 365))

Holt-Winters filtering

nts <- ts(daily$N, frequency = 7)

fit <- HoltWinters(nts, beta = FALSE, gamma = FALSE)

plot(fit)

fit <- HoltWinters(nts)

plot(fit)

library(forecast)

forecast(fit)

plot(forecast(HoltWinters(nts), 31))

Autoregressive Integrated Moving Average models

auto.arima(nts)

auto.arima(nts, approximation = FALSE)

plot(forecast(auto.arima(nts, D = 1, approximation = FALSE), 31))

Outlier detection

cts <- ts(daily$Cancelled)

fit <- auto.arima(cts)

auto.arima(cts)

library(tsoutliers)

outliers <- tso(cts, tsmethod = 'arima',

   args.tsmethod  = list(order = c(1, 1, 2)))

plot(outliers)

plot(tso(ts(daily$Cancelled)))

dfc <- as.data.frame(daily[, c('date', 'Cancelled'), with = FALSE])

library(AnomalyDetection)

AnomalyDetectionTs(dfc, plot = TRUE)$plot

More complex time-series objects

library(zoo)

zd <- zoo(daily[, -1, with = FALSE], daily[[1]])

plot(zd)

plot(cumsum(zd))