Machine Learning Anomaly Scoring - Comparative Reporting

Meter Solution Cloud Service (MSCS) includes productized comparative reporting for Machine Learning (ML)-based validation. The feature captures ML anomaly scores, relevant Validation, Editing, and Estimation (VEE) results, and inferred user actions, then presents aggregate statistics and detailed investigation results in the Meter ML and VEE Dashboard. It supports both historical evaluations of selected IMDs and ongoing production evaluation of ML exceptions.

MSCS provides comparative reporting that compares ML anomaly scores with VEE outcomes and corrective actions. In a non-production environment, you can evaluate already processed IMDs by cloning and scoring them with ML, then compare the results with VEE outcomes and the actions taken to correct VEE exceptions. In production, the Meter ML and VEE Dashboard uses statistics captured for IMDs that fail ML scoring to support ongoing evaluation and anomaly score threshold tuning.

This helps implementations tune ML anomaly score thresholds, investigate differences between ML and VEE outcomes, and continuously improve ML based validation.

Steps to enable and configure

To configure evaluation of historical IMDs with Machine Learning in a non-production environment:

  • Create an ML VEE exception type.

  • Update the MDM master configuration with the ML VEE exception type and set the anomaly score threshold to 0.

  • Configure a consumption extract type by using the ML Historical Verification Extract Type business object (D1-MLVerificationExtractType).

  • Configure the ML verification statistic control by using the ML Verification Control business object (D1-MLVerificationControl).

  • Create a D1-SGGMLPayloadProcessing entry to ingest ML scoring results.

  • Run these batch processes:

    • D1-GMLVI — Generate ML Historical Verification IMDs.

    • D1-EXHVI — Extract ML Historical Validation IMDs.

    • Wait for ML scoring results

    • D1-PLMLP — SGG ML Payload Processing Monitor Template.

    • D1-IMDML — IMD Monitor – Waiting for ML IMDs.

    • ML Verification Statistics Snapshot Monitor.

To configure ongoing production evaluation:

  • Create an ML VEE exception type using the D1-MLExceptionWithStats business object.

  • Update the MDM master configuration with the ML VEE exception and an appropriate fallback anomaly-score threshold.

  • Configure a consumption extract type by using the ML Historical Verification Extract Type business object (D1-MLVerificationExtractType).

  • Configure the ongoing verification statistic control by using the ML Ongoing Verification Control business object (D1-MLOngoingVerificationCntrl).

  • Schedule these batch processes:

    • D1-PLPSO — Payload Processing Monitor Template with Export.

    • Wait for ML scoring results

    • D1-PLMLP — SGG ML Payload Processing Monitor Template.

    • D1-IMDCS — Process IMD Control Stage Data.

    • D2-VEEDF — VEE Exception Monitor.

    • F1-STATS — Statistics Control Monitor.