AI-Generated vs Human Arduino Firmware Dataset: A Stylometric Approach for Detecting LLM-Generated Code

This dataset consists of 180 firmware source code samples for the Arduino platform, designed to distinguish between human-written and AI-generated firmware. The dataset is split into two categories: 90 human-written firmware samples and 90 AI-generated samples. The human-written code was sourced from official Arduino libraries, including examples that were committed prior to 2019, ensuring that no LLM-based generation tools were involved.

The AI-generated firmware samples were created using OpenAI's GPT-4o model, based on standardized prompts that describe specific Arduino projects. Each AI-generated firmware sample corresponds directly to a human-written sample, ensuring consistency in the structure and content of the code.