Fraud Detection — Live Dashboard

SinaApp real-time monitoring of 100K accounts

LIVE
Live POC — Fraud Detection
This agent-less POC is deployed on a single VPS without redundancy or High Availability (i.e. no active/active).
⚠️
Under Construction
Some services are currently unavailable.
The dashboard will resume automatically when services are restored.
Overview
Scenarios / min
—
execution rate
Accounts
—
Transactions
—
Fraud Detected
—
scenarios triggering
Clean
—
scenarios clean
Active Bursts
—
accounts with >5 txn/5min
Financial Impact
Sum of current balance for all 100K monitored accounts. Source: MongoDB accounts collection (live aggregate).
Total Balance
—
across all scanned accounts
Sum of current balance for accounts flagged as compromised (fraud detected by at least one scenario). Source: MongoDB — sum(balance) where compromised=true.
Amount Detected
—
balance of compromised accounts
Sum of blocked_balance — the account balance snapshot taken at the moment SinaApp blocked the account. Incoming transactions keep accumulating but outgoing transfers (drain) are prevented. Source: MongoDB — sum(blocked_balance) where blocked=true.
Amount Blocked
—
balance frozen at block time
Sum of outgoing transfers on accounts confirmed as compromised (true positives only — false positives are excluded since only fraud-level results contribute). Accumulated from Kafka scenario results. Sorry, what is gone is gone — but before SinaApp Behavioral Fraud Detection, you did not have a way to measure it.
Amount Drained
—
outgoing transfers on fraud accounts
Accounts currently blocked that are NOT compromised (false positives). These are automatically unblocked after a 2-minute cooldown. Amount shown should be deducted from Amount Blocked for the true protection figure. Source: MongoDB — sum(blocked_balance) where blocked=true and compromised=false.
False Positives
—
auto-released after 2min
True Positives / (True Positives + False Positives). How many of the blocked accounts are actually compromised.
Precision
—
blocked that are real fraud
True Positives / (True Positives + False Negatives). How many of the known compromised accounts have been detected and blocked.
Recall
—
fraud accounts caught
Harmonic mean of Precision and Recall: 2 × P × R / (P + R). Balanced measure of detection quality.
F1 Score
—
harmonic mean P & R
TP: blocked + compromised. FP: blocked + NOT compromised. FN: NOT blocked + compromised (missed fraud). TN: NOT blocked + NOT compromised.
Confusion Matrix
—
TP / FP / FN / TN
Scenario Status (cycle 30s)
Status Scenario Level Elapsed Checks Fraud Clean Blocked Amount Blocked Amount Drained Last Check Code
Loading…
Recent History
Time Scenario Level Elapsed
Loading…