anomalies: Anomaly Indentification

In chrishaarstick/madtstools: MadStork Time Series Tools

Description

Function to idendtify anomalies in time series data. Function similar to Seasonal ESD approach here : https://arxiv.org/pdf/1704.07706.pdf

Usage

anomalies(df, index_var = NULL, x_var = NULL, ts_frequency = 1,
  stl_args = list(s.window = "periodic"), confidence = 0.99,
  direction = "both")

Arguments

df	input data.frame
index_var	optional column name of index variable. default is NULL which uses the first column in the input data.frame
x_var	optional column name of numeric variable to analyze. efault is NULL which uses the second column in the input data.frame
ts_frequency	time series frequency. default is 1
stl_args	list of arguments to pass to stl function. s.window argument required - default is 'periodic'
confidence	anomaly threshold confidence level. higher levels increase the anomaly threshold
direction	direction of anomaly idendification. Options are 'positive', 'negative', or 'both'

Value

data.frame with input dataset, stl components and flag for anomalies

Examples

library(tibble)
set.seed(319)
 n <- 100
  df <- tibble(index = 1:n,
               x = as.numeric(arima.sim(model = list(0,0,0), n),
                                        rand.gen = function(x) rt(x, 4)))

anoms <- anomalies(df, "index", "x", ts_frequency = 5, 
                   confidence = 0.95, direction = "both")

chrishaarstick/madtstools documentation built on May 28, 2019, 11:04 p.m.