AirTagged Dataset and Processing Framework for IoT

This paper is a technical report that presents an open-source dataset and processing framework for Bluetooth Low Energy (BLE) advertisement packets, specifically targeting personal tracking devices in high-density Internet of Things (IoT) environments. The dataset comprises 13 million labeled BLE packets collected over 200 hours, aimed at facilitating machine learning (ML) applications for device detection and classification. The report outlines the need for a comprehensive dataset to address privacy concerns associated with BLE trackers, such as Apple's AirTag, which can be misused for stalking. It introduces a modular Task-Group Framework designed for efficient data preprocessing, feature extraction, and dynamic visualization of BLE packets. The paper also discusses the dataset's collection methodology, the structure of BLE advertisement packets, and the importance of including diverse device states to enhance model training. By providing these resources, the report aims to support the development of vendor-agnostic, privacy-preserving ML solutions for detecting unauthorized BLE tracking in various real-world scenarios.