What causes ML_ANOMALY_DETECTION_FAILED?

The training series contains fewer data points than the minimum required for the model to learn a baseline. The timestamp column has gaps or irregular intervals that prevent seasonality detection. The prediction series uses a different series key or column name than was used during training. The anomaly detection model instance was dropped and not recreated before prediction was called

How do I fix ML_ANOMALY_DETECTION_FAILED?

Step 1: Confirm the model instance exists using SHOW SNOWFLAKE.ML.ANOMALY_DETECTION in the target schema.. Step 2: Review the training data — ensure the time series has sufficient rows (at minimum several weeks of data for daily data) and no large gaps.. Step 3: Verify the series_colname, timestamp_colname, and target_colname match exactly between training and prediction calls.. Step 4: Retrain the model using SNOWFLAKE.ML.ANOMALY_DETECTION if the model was dropped or training data has changed significantly.. Step 5: Test predictions on a small recent slice of data before running the full production inference pipeline.

High severityexecution

Power BI Refresh Error:
ML_ANOMALY_DETECTION_FAILED

What does this error mean?

A Snowflake ML Anomaly Detection model failed to generate predictions, typically because the training data was too sparse, the timestamp column is incorrect, or the model instance was dropped.

Common causes

1The training series contains fewer data points than the minimum required for the model to learn a baseline
2The timestamp column has gaps or irregular intervals that prevent seasonality detection
3The prediction series uses a different series key or column name than was used during training
4The anomaly detection model instance was dropped and not recreated before prediction was called

How to fix it

1Step 1: Confirm the model instance exists using SHOW SNOWFLAKE.ML.ANOMALY_DETECTION in the target schema.
2Step 2: Review the training data — ensure the time series has sufficient rows (at minimum several weeks of data for daily data) and no large gaps.
3Step 3: Verify the series_colname, timestamp_colname, and target_colname match exactly between training and prediction calls.
4Step 4: Retrain the model using SNOWFLAKE.ML.ANOMALY_DETECTION if the model was dropped or training data has changed significantly.
5Step 5: Test predictions on a small recent slice of data before running the full production inference pipeline.

Frequently asked questions

How much training data does Snowflake ML Anomaly Detection require?

The minimum is roughly 30 observations at the prediction frequency. For daily data, 30 days is the absolute minimum; Snowflake recommends at least 90 days for reliable seasonality detection.

Can the anomaly detection model handle multiple series simultaneously?

Yes. When training with a series_colname column, the model learns per-series baselines. At prediction time, the same series keys must be present or predictions for missing series will not be generated.