ML wireless eavesdropper detection

• Shubham Banwal

• August 2025

Machine Learning–Based Eavesdropping Detection in Wireless Communication Systems

Problem:

Wireless communication systems are vulnerable to eavesdropping attacks, where unauthorized entities intercept transmitted data. Detecting such threats in real time is challenging due to signal variability, noise, and lack of labeled attack data.

Approach:

Designed and implemented a detection system using both supervised and unsupervised machine learning models to identify anomalies and potential eavesdropping behavior in wireless communication environments.

System Design:

• Data acquisition and preprocessing pipeline for wireless signal features
• Feature extraction layer for identifying patterns in communication signals
• Supervised learning models for classification of normal vs compromised signals
• Unsupervised models for anomaly detection in unknown attack scenarios
• Evaluation framework for model accuracy and detection performance

Key Contributions:

• Developed machine learning pipelines for processing and analysing wireless communication data
• Implemented supervised models (e.g., Logistic Regression, ANN) for classification tasks
• Applied unsupervised techniques (e.g., One-Class SVM) for anomaly and intrusion detection
• Compared performance of models across different detection scenarios
• Generated interpretable outputs to support decision-making in security contexts
• Evaluated system robustness under noisy and variable signal conditions

Constraints & Tradeoffs:

• Limited labeled data for supervised learning
• High variability and noise in wireless signal environments
• Tradeoff between detection sensitivity and false positives
• Balancing model complexity with computational efficiency