Detect Seasonality Changes with Automatic Threshold

Detects points where seasonality starts or stops, using automatic threshold selection instead of a fixed value.

Usage

detect.seasonality.changes.auto(
  fdataobj,
  period,
  threshold_method = "otsu",
  threshold_value = NULL,
  window_size = NULL,
  min_duration = NULL
)

Arguments

fdataobj

An fdata object.

period

Known seasonal period.

threshold_method

Method for threshold selection:

"fixed"

Use threshold_value as fixed threshold

"percentile"

Use threshold_value as percentile of strength distribution

"otsu"

Use Otsu's method for bimodal separation (default)

threshold_value

Value for "fixed" or "percentile" methods.

window_size

Width of sliding window for strength estimation. Default: 2 * period.

min_duration

Minimum duration to confirm a change. Default: period.

Value

A list with components:

change_points

Data frame with time, type, strength_before, strength_after

strength_curve

Time-varying seasonal strength used for detection

computed_threshold

The threshold that was computed/used

Details

Otsu's method automatically finds the optimal threshold for separating seasonal from non-seasonal regions based on the strength distribution. This is particularly useful when you don't know the appropriate threshold for your data.

Examples

# Signal that transitions from seasonal to non-seasonal
t <- seq(0, 20, length.out = 400)
X <- ifelse(t < 10, sin(2 * pi * t / 2), rnorm(sum(t >= 10), sd = 0.3))
X <- matrix(X, nrow = 1)
fd <- fdata(X, argvals = t)

# Detect changes with Otsu threshold
changes <- detect.seasonality.changes.auto(fd, period = 2)
print(changes$computed_threshold)
#> [1] 0.516327