S
F
Fraud Detection · Big Data Processing Layer

Spark Batch + Flink Stream Dashboard

Pipeline Active
—
Overview
Spark Batch
Flink Stream
Alerts
Spark Processed
0
batch transactions
Flink Processed
stream events
Critical Alerts
0
flink rule engine
High Risk
0
score ≥ 0.5
Avg Latency
—
flink p50 ms
Fraud Rate
0%
detected / total
Processing Throughput Live
Spark Batch (tx/min)
Flink Stream (event/s)
Risk Distribution Flink
Critical0
High0
Medium0
Low0
Real-Time Data Flow Topology
CSV transactions.csv kafka_producer docker-compose 容器 Kafka transactions.raw Flink Rule Engine data-api :8000 Kafka Consumer Kafka Alerts transactions.alerts DataPipeline graph_builder → HeteroData GNN API :8001 GraphSAGE 推理 train.py ⏰ Retrain 每週一 3:00 → best_model.pt Dashboard :8081 Demo Frontend Spark ⏰ Spark DAG 每天 2:30 Parquet spark/api.py :8092 best_model.pt 載入
Recent Alert Timeline Flink
Waiting for Flink alerts…
Total Transactions
0
batch processed
Customer Nodes
0
feature vectors
Merchant Nodes
0
feature vectors
Edge Count
0
graph relationships
Fraud Rate
0%
detected in batch
Transaction Type Distribution Spark
Graph Edge Type Distribution Spark
Fraud Rate by Time Step Spark
Fraud Count
Fraud Rate (%)
Top Suspicious Customers (Spark Features) Spark
Customer IDTotal SentTotal Received Outgoing TxTransfer RatioCashOut RatioFraud
Total Alerts
0
CRITICAL + HIGH
Critical
0
immediate action
High
0
investigation needed
Top Signal
—
most triggered rule
Alert Feed — Flink Rule Engine Output Real-Time
TimeRiskSenderReceiver AmountTypeRisk ScoreSignals
No alerts yet