📞 Real-Time Call Screening & Scam Detection System
I. Executive Summary
Current anti-scam measures mostly focus on the victim side (blocking or warning), while the source-side abnormal calling behavior has long been left unmonitored. This proposal introduces a system based on abnormal call-behavior analysis that detects suspicious scam calls through telecom metadata (not call content). It enables early interception of high-risk phone numbers and allows telecom operators and law enforcement to accurately trace scam origins.
II. Problem Analysis
- Existing anti-scam systems rely heavily on complaints or post-victim reports, which is passive.
- Scam groups use high-volume SIM cards and automated dialing systems, making anomalies hard to detect quickly.
- Legitimate outbound business calls (insurance, customer service, etc.) often look similar to scam calls, causing high misidentification rates.
- There is no behavioral-level fraud detection and no enterprise whitelist mechanism in current systems.
III. Solution
Establish a “Call Behavior Screening + Whitelist System” that analyzes call metadata (time, region, frequency, duration) to determine abnormal patterns. This model does not require monitoring call content; it works solely through call metadata, protecting privacy while improving efficiency.
(1) Behavioral Detection Model
| Indicator | Normal Range | Abnormal Range | Explanation |
|---|
| Daily Call Volume | < 100 calls | > 300 calls | High-volume batch calling suggests scams |
| Geographic Pattern | Same/nearby region | No regional correlation | Random region dialing with no commercial logic |
| Avg. Call Duration | 3–10 mins | <30 sec or >20 min repeatedly | Short = probing; Long = manipulation / persuasion |
| Target Repetition | Repeated customers | Almost all unique | Randomized number blasting |
| Time Distribution | Mostly business hours | All day or late night | Automated or offshore dialing |
(2) Whitelist System
- Companies register outbound call usage (name, purpose, contact person).
- Monthly activity reports used for verification and authentication.
- Legitimate outbound units (insurance, customer service, etc.) are excluded from false flags.
(3) Risk Scoring Model
| Score | Risk Level | Action |
|---|
| 0–40 | Normal | No action |
| 41–70 | Observation | Monitored |
| 71–100 | High risk | Tracked & flagged as potential scam origin |
IV. Implementation Steps
- Data Integration: telecom provides anonymized call logs (dial time, region, duration).
- Behavioral Model Training: past scam number samples are used to train anomaly detection.
- Whitelist Registration: legitimate outbound entities join registry to prevent misclassification.
- Automatic Tagging & Alerts: daily generation of “high-risk number lists.”
- Law Enforcement Integration: connect with Anti-Fraud Center for tracing and enforcement.
V. Expected Benefits
| Benefit | Description |
|---|
| 🎯 Source Interception | Detects scam numbers before mass victimization occurs |
| 🔒 Privacy Protection | Relies on metadata only; no wiretapping or content analysis |
| 🧠 Smart Adaptation | Algorithm evolves with new scam techniques |
| 💼 Public–Private Collaboration | Telecom + Government + Enterprises co-build prevention ecosystem |
| 💰 Low Cost | Uses existing telecom logs; minimal deployment overhead |
VI. Extended Applications
- Build a Nationwide Scam Call Risk Map visualizing hot zones.
- Mobile devices display “Suspected Scam Number” alerts in real time.
- Integration with banking hotlines and digital anti-fraud apps.
- Provide open data access for public transparency.
