spectralAnomaly spectralAnomaly website

CRAN status CRAN checks R-CMD-check Dependencies Downloads Codecov test coverage

The {spectralAnomaly} package is a simple set of tools for R users to detect anomalies in data, such as a time series, using the ‘Spectral Residual’ method.

Installation

The {spectralAnomaly} can be installed directly from CRAN:

install.packages('spectralAnomaly')

You can also install the latest development version of {spectralAnomaly} like so:

remotes::install_github('al-obrien/spectralAnomaly')

Examples

library(spectralAnomaly)

Create a noisy time series with a modest anomaly threshold

test_ts <- ts(rnorm(12*6,10,2), start=c(2009, 1), end=c(2014, 12), frequency=12)
ts_scores <- anomaly_score(test_ts, score_window = 25)
plot(test_ts, type = 'l')
points(test_ts, col = ifelse(ts_scores > quantile(ts_scores, prob = 0.95),'red',NA), pch = 16)

Anomalies in a noisy time series.

Create a series with a step

test_ts_step <- c(rnorm(1, 1, n=250),
                  rnorm(10, 1, n=250))
ts_scores <- anomaly_score(test_ts_step, score_window = 100)
plot(test_ts_step, type = 'l')
points(test_ts_step, col = ifelse(ts_scores > quantile(ts_scores, prob = 0.99),'red',NA), pch = 16)

Anomalies in a time series with a breakpoint.

References

  1. Time-Series Anomaly Detection Service at Microsoft
  2. ML-based Anomaly Detection
  3. Saliency Detection
  4. anomalydetector