Back to all work
Research · Cal Poly Pomona

IoT Network Security

IoT Security Analytics Using Packet Capture (PCAP) Files · Faculty advisor: Dr. Mohammad Husain · May 2024 – Oct 2024

I studied cybersecurity threats in IoT networks (smart cameras, voice assistants, connected appliances) by training machine-learning classifiers on packet-capture (PCAP) data to detect malicious network activity, benchmarking five algorithms on accuracy, runtime, CPU, and memory across millions of samples.

Network SecurityMachine LearningData AnalysisTechnical Communication
Contributions

What I did

  • Built the experiment pipeline in Python / scikit-learn: extracting and sorting PCAP data into attack vs. benign sets, shuffling for unbiased training, and benchmarking classifiers on accuracy, runtime, CPU, and memory.
  • Compared five algorithms: Decision Tree, Random Forest, Logistic Regression, linear SVC, and GaussianNB, across millions of samples.
  • Found Decision Tree most accurate at flagging suspicious traffic (Random Forest close behind) at the cost of runtime; Logistic Regression and linear SVC won on medium-sized data, and GaussianNB traded precision for speed.
  • Presented the work as a research poster under faculty advisor Dr. Mohammad Husain.
Stack

Tools

Pythonscikit-learnpandas · NumPymatplotlib · seabornPCAP AnalysisML Benchmarking
Publication

Research poster

The full poster: problem, method, benchmark results, and conclusions. Tap to enlarge, or grab the PDF.