// AML · Anomaly Detection Engine

Transaction Risk Screening

A rule-based anomaly engine modelled on Isolation Forest logic — trained on 10,000 synthetic SG remittance transactions. Flags structuring, velocity abuse, blacklisted receivers, and unusual behavioural patterns.

sklearn.IsolationForestn_estimators=100contamination=0.05max_samples=2568 trees extracted
// Preset Scenarios
// Transaction Parameters
00:0012:0023:00
010018RISK SCORE
LOW RISKIF score: 0.5732
Proceed — standard monitoring applies
TRIGGERED SIGNALS (0)
No anomalies detected
Transaction profile matches normal corridor behaviour
// Session Pattern Analysis (0 transactions)
TimeAmountTypeReceiverRisk ScoreLevel
No transactions analysed yet. Click "Analyse Transaction" to add to session log.
// Methodology

The anomaly score is computed by traversing 8 pre-extracted Isolation Forest trees. Each tree was trained on 256 randomly sampled transactions from the synthetic dataset. The expected path length E[h(x)] across all trees is normalised by c(n) = 2H(n−1) − 2(n−1)/n where n = 256, yielding a score in (0, 1) where values above 0.6 indicate increasing anomalousness.

The model score is augmented with a rule-based engine that mirrors the first-generation transaction monitoring systems used by MAS-regulated banks: blacklist screening, threshold-proximity detection, velocity counting, temporal analysis, crypto wallet routing detection, and receiver identifier entropy scoring. The combined score drives four risk tiers (LOW / MEDIUM / HIGH / CRITICAL) consistent with MAS Notice PSN02 AML/CFT guidance for digital payment tokens.