Low-cost, edge-based system that detects voice phishing (vishing) attacks in real-time using on-device keyword scoring — no cloud dependency, full data privacy. Published as a peer-reviewed research paper.
5-layer architecture: audio acquisition → preprocessing → fraud analysis → decision → multi-modal alert
| Metric | Value |
|---|---|
| Accuracy | 95.6% |
| Precision | 94.8% |
| Recall | 96.2% |
| F1-Score | 95.5% |
| Detection Latency | 320ms |
| False Alarm Rate | 4.1% |
| Cost | ₹3,500 |
The system captures voice input, converts it to processable tokens, and compares them against a weighted keyword database (OTP, lottery, urgent, verification, etc.). A cumulative fraud score S = Σ(wᵢxᵢ) is computed — if it crosses the threshold, the system triggers dual alerts via GSM SMS and LCD display, entirely on-device.
| Component | Purpose |
|---|---|
| STM32 (32-bit MCU) | Main processing unit |
| Microphone module | Voice acquisition |
| 16x2/20x4 LCD | Warning display |
| Buzzer | Audio alert |
| SIM800L (GSM) | SMS alerts |
| ESP32 | Bluetooth connectivity (future scope) |
STM32 · Embedded C · GSM (SIM800L) · Python (TF-IDF + Logistic Regression evaluation)
firmware/ → STM32 embedded C code
analysis/ → Fraud scoring evaluation (Python)
docs/ → Paper, figures, architecture diagrams
legacy/ → Earlier ESP32 + vibration-motor prototype
Jahnvi P. Issar, Irfa Arshad, Mansi Chate — Dept. of E&TC, Dr. D. Y. Patil Institute of Technology, Pune