Anomaly Detection Math Settings

Every AD model consists of an underlying forecasting model that tries to forecast what should happen under normal circumstances. On top of that, an anomaly detection layer is built to work with the residuals of reality and these forecasts. In this section, you can influence parameters of this layer to ensure that it really learns what is anomalous. By default, TIM tries to do everything automatically so you do not have to change anything, but sometimes it might be useful.

Where to set this in TIM connector and API


Where to set this in TIM Studio


Data Normalization

When normalization is on, features entering the abnormal behavior modelling are descaled by mean and standard deviation. If not provided, a default value will be used.

dataNormalization: true

Model Complexity

Maximal complexity to search when building the abnormal behavior model. You need to provide explicit number for exact model complexity selection. Defined range: 1 – 30. If not specified, TIM Engine will determine optimal model complexity automatically.

maxModelComplexity: 20

Detection Features

Features for anomaly detection. An arbitrary combination of them can be used, however, at least one has to be selected. More about features here

features: [Residual, ResidualChange, Fluctuation, FluctuationChange, Imbalance, ImbalanceChange, Frequency, FrequencyChange, Hour, DayOfWeek, Month]