EX

Enterprise AI Administration

Revenue Leakage Detection Model

Model Details · Telecom AI Model Registry

Model Details

Revenue Leakage Detection Model

Algorithm: Isolation Forest + Business RulesAPPROVED DESIGN
Code
MDL-LEAK-DETECT
Version
v1.4
Owner
Head of Revenue Assurance
Governance owner
Model Risk Committee

Governance configuration shown for review and validation. Read-only.

Business purpose

Flags billing, rating and interconnect records that indicate unbilled or under-billed usage.

Production-recommended design specification. This record does not mean the model has been trained, fitted, deployed or validated.

Technical specification

Model family
Unsupervised Anomaly Detection + Business Rules
Algorithm
Isolation Forest + Business Rules
Implementation library
IsolationForest + RA rule set
Model type
Hybrid Anomaly Detection
Explainability method
Anomaly score + triggered rules + feature contribution where supported

Implementation governance

Governance status
APPROVED DESIGN
Implementation status
MODEL DEVELOPMENT REQUIRED
Validation status
VALIDATION REQUIRED
Lifecycle stage
Design
Current performance
Not yet measured
Refresh cadence
Daily
Last reviewed
30 Jul 2026
Next review due
30 Oct 2026
Historical sample reference
Confirmed-case rate 0.74 — earlier sample figure, kept for provenance only; not a validated model result.

Business

Mission area
Fraud & Revenue Assurance
Decision journey
fraud-revenue-assurance
Target variable
leakage_case_flag
Output
Leakage exception list with estimated amount

Input features

Governed variables, in feature-rank order

  1. 1.usage_pattern_deviationDeviation of current usage from the subscriber's 90-day profile (z-score).
  2. 2.duplicate_recharge_countRecharges reusing a voucher or repeating within 60 seconds, last 7 days.
  3. 3.fraud_anomaly_scoreUnsupervised anomaly score on calling and recharge behaviour (0–1).

Related Intelligence

Linked governed records

Decision journey