Changepoint

YAML type: changepoint | Category: Adaptive

Detects structural level shifts using ML.NET changepoint analysis (IID or SSA variants).

Description

Changepoint finds lasting structural breaks. Each YAML block sets variant (iid or ssa) and processing (none, rm, cs, rv). Series need more than 12 points.

How it works

Changepoint illustration

Changepoint looks for lasting structural breaks using ML.NET analysis. Optional preprocessing (none, rolling mean, cumsum, or rolling variance) shapes the signal; variant chooses IID or SSA (seasonal) detection.

When confidence and history-length criteria are met, a ChangePoint anomaly marks the break. Series need more than about twelve points. Multiple changepoint blocks can run with different variant/processing combinations.

Setup

Start with iid + none. Use SSA when seasonality matters. Preprocessing (rm rolling mean, cs cumsum, rv rolling variance) shapes what 'change' means.

Configure detectors on the data view Analysis tab or in YAML. Validate with Data Preview before enabling production schedules.

Settings

Parameter	Default	Description
`variant`	`iid`	Detection variant -- see Allowed values
`processing`	`none`	Preprocessing -- see Allowed values
`confidence`	`95`	Detection confidence
`changeHistoryLength`	`30`	History length for change detection
`windowLength`	`15`	For `rm` / `rv`
`seasonalityWindowSize`	`7`	For `ssa`

Allowed values

`variant`

Value	Meaning
`iid` (default)	Independent / IID changepoint detection
`ssa`	Seasonal SSA changepoint detection

`processing`

Preprocessing applied before changepoint analysis.

Value	Meaning
`none` (default)	No preprocessing (aliases also accepted: `raw`, `direct`, `identity`, `ziadne`, `ziande`)
`rm`	Rolling mean -- requires `windowLength`
`cs`	Cumulative sum
`rv`	Rolling variance -- requires `windowLength`

seasonalityWindowSize applies when variant is ssa.

Shared series filters

Most detectors also accept optional series filters in params:

Parameter	Purpose
`maxGapAmount`	Skip series when gap amount is too high (`null` = no limit). Catalog create-default is often `5`.
`minValue`	Skip series whose average value is below this (`null` = no limit). Catalog create-default is often `1000`.
`maxValue`	Skip series whose average value is above this (optional; omit/`null` = no limit)
`useSpecialDates`	When `true`, suppress anomalies that overlap special dates

When a filter skips a series, Data Preview shows the reason.

Configuration example

analysis:
  detectors:
    - type: changepoint
      params:
        variant: iid
        processing: none
        confidence: 95
        changeHistoryLength: 30
    - type: changepoint
      params:
        variant: ssa
        processing: rm
        confidence: 95
        changeHistoryLength: 30
        windowLength: 15
        seasonalityWindowSize: 7

Using the detector

Treat as a structural-break signal. Follow up with MoM/WoW or comments to explain the business change.

Use cases

After major product launches
Pricing or policy changes
Detecting variance regime shifts (rv)

Specifics

All variants store anomaly type ChangePoint. Multiple changepoint blocks can coexist.