ML-03 · Compliance & Resilience · Payment Volume Surveillance · Time-Series

Time-Series Anomaly Detector

Rolling-window z-score and STL-style seasonal decomposition anomaly detection on synthetic payment volume time series. Produces trend/seasonal/residual decomposition, a control chart with UCL/LCL (3σ bounds), anomaly flag table with severity classification, and a naïve 30-day ARIMA-lite forecast. Chains from SIM-03 (Basel RWA Scenario Modeler) — stressed capital scenarios can flag unusual payment flow patterns during systemic events. Feeds RCA-01 (FRTB IMA Pre-Validator). All computation client-side — no payment data transmitted.

detect_timeseries_anomalies risk_control DORA Art. 17 Monitoring EBA GL/2021/03 PSD2 Art. 96 Fraud Reporting FCA SYSC 8
Educational / Pre-validation only. Synthetic payment volume time series generated client-side. Anomaly detection uses rolling z-score and simplified seasonal decomposition — not a production-grade surveillance system. Chains from SIM-03 to import capital-stress context. For real payment fraud surveillance, see also ML-01 (Isolation Forest). No data transmitted.
Preset 1
Daily Retail Payments
365 days · weekly seasonality · 3 injected spikes
Preset 2
Weekly Settlement
104 weeks · monthly cycle · EOM spikes
Preset 3
Hourly Real-Time
720 hours · intraday pattern · stress scenario
Chain Intake — paste AP2 JSON from SIM-03 (Basel RWA Scenario Modeler)