AI Real-Time WLAN Threat Detection
Budget: $30 – $250 USD
I need a working proof-of-concept that hardens my wireless network by analysing live traffic and flagging hostile activity the moment it appears. The model must reliably catch the attacks that worry me most: Denial of Service (DoS), Man-in-the-Middle (MitM) intrusions, classic packet sniffing, and ARP spoofing. Everything has to run in real time on a modest on-prem machine, so efficient feature extraction and lightweight inference are a priority.
The task breaks down naturally into three parts. First, capture and label representative WLAN traffic—public datasets are fine as a starting point, but I also want a small tool that lets me pipe raw pcap streams into the training set so the system can learn my network's quirks. Second, build and train the detection engine: a well-commented Python project that leverages scapy, pandas, and the scikit-learn implementation of Random Forest, which is my preferred choice for its balance of performance and interpretability. Third, wrap the model in a daemon that watches the interface, raises an alert (syslog plus a webhook) when an attack pattern is scored above the chosen threshold, and writes a short JSON report for later forensics.
Deliverables
Source code and requirements.txt/poetry.lock
Trained Random Forest model file and reproducible training script
Small CLI utility to feed live or recorded pcap into the detector
README that explains setup, expected performance, and how to extend with new attack types
Acceptance criteria
At least 90% detection accuracy on a held-out test set for the four attack classes listed above
Average inference time under 100 ms per packet on a standard laptop CPU
Clean, PEP-8 compliant code with inline comments
Provide any clarification questions early so we can keep the iteration cycle tight.
The task breaks down naturally into three parts. First, capture and label representative WLAN traffic—public datasets are fine as a starting point, but I also want a small tool that lets me pipe raw pcap streams into the training set so the system can learn my network's quirks. Second, build and train the detection engine: a well-commented Python project that leverages scapy, pandas, and the scikit-learn implementation of Random Forest, which is my preferred choice for its balance of performance and interpretability. Third, wrap the model in a daemon that watches the interface, raises an alert (syslog plus a webhook) when an attack pattern is scored above the chosen threshold, and writes a short JSON report for later forensics.
Deliverables
Source code and requirements.txt/poetry.lock
Trained Random Forest model file and reproducible training script
Small CLI utility to feed live or recorded pcap into the detector
README that explains setup, expected performance, and how to extend with new attack types
Acceptance criteria
At least 90% detection accuracy on a held-out test set for the four attack classes listed above
Average inference time under 100 ms per packet on a standard laptop CPU
Clean, PEP-8 compliant code with inline comments
Provide any clarification questions early so we can keep the iteration cycle tight.