Anomaly detection in a network based on a key performance indicator prediction model
摘要
說明書
The KPI prediction model may be configured in a training state until the KPI prediction model satisfies a training-related threshold. In some implementations, the KPI prediction model may be configured in a training state until a specified quantity of measured values for the KPI have been used for training the KPI prediction model. In some implementations, the KPI prediction model may be configured in a training state until a specified amount of time has passed during which the KPI prediction model receives measured values for the KPI. In some implementations, the KPI prediction model may be configured in a training state until the KPI prediction model satisfies a prediction accuracy threshold. When in a training state, the KPI prediction model may not attempt to identify an anomaly.
In some implementations, the KPI prediction model may determine a prediction for particular values of the KPI using historical values of the KPI that have one or more parameters that match one or more parameters of the particular values of the KPI. For example, the KPI prediction model may use only historical values of the KPI that were measured during the same period of time in a day and on the same day of the week as a particular value of the KPI that is to be predicted. In another example, the KPI prediction model may use only historical values of the KPI that were measured on the same holiday as the particular value of the KPI that is to be predicted.
