Skip to contents

Regime Detection for Time Series Data

Source: R/regimes.R
detect_regimes.Rd

Detects regime changes in time series data using multiple methods including cumulative peaks, changepoint detection, variance shifts, threshold analysis, gradient changes, and entropy analysis.

Usage

detect_regimes(
  data,
  method = "smart",
  sensitivity = "medium",
  min_change,
  window = 10L,
  peak = 2,
  cumulative = 0.6
)

Arguments

data

[ts, numeric()]
Univariate time series data.

method

[character(1): "smart"]
Regime detection method. The available options are:

  • "cumulative_peaks": Detects cumulative complexity peaks using Z-tests.

  • "changepoint": Change point detection (multi-window mean-shift test).

  • "threshold": Adaptive quartile-based regime classification.

  • "variance_shift": Detects changes in variance patterns.

  • "slope": Detects changes in local slope (rolling linear models).

  • "entropy": Detects changes in the Shannon entropy of the complexity series, calculated in rolling windows.

  • "smart" (default): Combines gradient, peaks, and changepoint methods.

  • "all": Applies all individual methods listed above and uses ensemble voting.

sensitivity

[character(1): "medium"]
Detection sensitivity level. The available options are: "low", "medium", and "high". Controls thresholds and window sizes within the detection methods.

min_change

[integer(1)]
Minimum number of observations between changes. If not provided, the value is determined automatically (typically 10% of observations, minimum of 10).

window

[integer(1): 10L]
Base window size for rolling calculations. This is further adjusted by sensitivity

peak

[numeric(1): 2.0]
Base z-score threshold for individual peak detection with the "cumulative_peaks" method. Adjusted by sensitivity.

cumulative

[numeric(1): 0.6]
A value between 0 and 1 that defines the base proportion threshold for identifying cumulative peak regions. Adjusted by sensitivity.

Value

An object of class regimes which is a tibble containing the following columns:

value: Original time series data. time: Original time points. change: A logical vector indicating regime changes. id: An integer regime identifier. type: Type of change detected by the method. magnitude: Magnitude of the change (method-specific interpretation) confidence: Confidence in the detection (method-specific interpretation, typically between 0 and 1, or NA) stability: Categorical stability: "Stable", "Transitional", and "Unstable". score: A numeric stability score between 0 and 1.

Examples

set.seed(123)
ts_data <- stats::arima.sim(list(order = c(1, 1, 0), ar = 0.6), n = 200)
regimes <- detect_regimes(
  data = ts_data,
  method = "threshold",
  sensitivity = "medium"
)