Smart Home Traffic Shaping for Privacy

ISEF Category: Systems Software

Subcategory: Cybersecurity  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Your smart speaker, camera, and light bulb may leak more than you think. Even when content stays encrypted, packet size and timing can still reveal what each device is doing. That means an ISP or attacker can fingerprint your home from traffic patterns alone. You can test whether a router-side shaper can hide those clues.

What Is It?

This project studies whether a router can make smart-home network traffic harder to fingerprint. Fingerprinting means identifying a device by patterns in its packets, like packet size, burst timing, and idle gaps. Think of it like recognizing a person by their walk, even if you cannot see their face.

A traffic shaper changes how packets leave the router. It may add padding, delay packets, or regularize bursts so the traffic looks less distinctive. Your goal is to see whether those changes reduce the accuracy of public IoT device classifiers. If the classifier starts guessing wrong, the privacy layer is doing its job.

Why This Is a Good Topic

This makes a strong science fair topic because you can test it with real data, clear metrics, and repeatable comparisons. You are not just building code, you are measuring privacy gain against performance cost, such as extra delay or added bandwidth. The topic connects to a real problem, since smart homes keep growing and traffic metadata can expose private behavior. You can also learn traffic analysis, basic machine learning evaluation, and experimental design.

Research Questions

• How does adversarial padding change the accuracy of an IoT device classifier on public traffic traces?
• What is the effect of packet timing randomization on device identification accuracy?
• Does fixed-rate shaping reduce fingerprinting more than burst-based shaping?
• To what extent does padding increase total bandwidth overhead for common smart-home traffic patterns?
• Which traffic features remain most predictive after shaping, packet size, interarrival time, or burst length?
• How does the shaper affect classification accuracy across different device types, such as cameras, speakers, and plugs?

Basic Materials

• Raspberry Pi or similar single-board computer with Ethernet port.
• Home router or spare router that supports custom firmware or traffic redirection.
• Linux laptop for setup and analysis.
• Ethernet cables.
• Wireshark for packet inspection.
• Python installed on a laptop or desktop.
• Public IoT traffic dataset, such as the UNSW IoT dataset.
• External storage for packet captures and logs.
• Digital stopwatch or system logs for timing checks.

Advanced Materials

• Raspberry Pi or mini PC running Linux as a router.
• Managed switch or network tap for cleaner capture setup.
• Multiple smart-home devices, or device traffic traces from a lab.
• Packet capture adapter and Ethernet cables.
• OpenWrt, Linux tc, or nftables for shaping rules.
• Python packages for analysis and machine learning evaluation.
• Jupyter Notebook for experiments and plots.
• Public IoT traffic datasets for baseline and validation.
• NAS or secure storage for large trace files.

Software & Tools

• Wireshark: Captures and inspects packet size, timing, and flow patterns.
• Python: Processes traces, computes metrics, and compares classifier results.
• Jupyter Notebook: Organizes analysis, plots, and experiment notes in one place.
• scikit-learn: Trains and scores baseline classifiers on traffic features.
• OpenWrt: Provides a router platform for testing shaping rules on real hardware.

Experiment Steps

1. Define the privacy signal you want to hide, such as packet size patterns, timing bursts, or both.
2. Choose one traffic-shaping method to test first, and set a baseline with no shaping.
3. Map the features a public classifier uses, so you know what your defense needs to disrupt.
4. Design controls that separate privacy gain from performance cost, such as bandwidth overhead and latency.
5. Plan how you will score success, using classifier accuracy, confusion matrices, and overhead metrics.
6. Decide whether you will test live device traffic, public traces, or both, then keep the pipeline consistent across conditions.

Common Pitfalls

• Training and testing on the same traffic trace, which makes the classifier look weaker or stronger than it really is.
• Changing router settings between runs, which adds noise and hides the effect of your shaper.
• Measuring only accuracy, which misses bandwidth overhead and delay tradeoffs.
• Using a dataset with one device family, which makes your results too narrow to generalize.
• Capturing traffic on a busy home network, which mixes unrelated packets into your trace.

What Makes This Competitive

A strong project goes beyond showing that shaping lowers classifier accuracy. You can compare multiple defenses, measure both privacy and overhead, and test whether the method still works across device types and feature sets. Strong entries also use clean train-test splits, confusion matrices, and statistical comparisons instead of one single accuracy number. If you can show which traffic features survive the defense, your project starts to look like real systems research.

Project Variations

• Test the same shaper on traffic from cameras, plugs, and speakers to see whether device type changes the privacy gain.
• Compare adversarial padding with fixed-rate shaping to measure which method gives better privacy for less overhead.
• Analyze whether a defense trained on one dataset still hides fingerprints in a different public IoT trace set.

Learn More

• MIT OpenCourseWare: Search for networking, security, and computer systems courses that cover packet processing and traffic analysis.
• PubMed: Search for review articles on privacy risks from smart-home network traffic and metadata leakage.
• arXiv: Search for recent preprints on IoT traffic classification and traffic shaping defenses.
• USENIX Security Proceedings: Search for peer-reviewed papers on network traffic analysis and privacy-preserving defenses.
• NIST: Look up cybersecurity guidance and terminology for secure network design and risk framing.