Detecting scraper bots through scroll behaviour

The author applied burstiness (B) and memory (M) metrics to JavaScript scroll-event timings from the FP-Agent dataset, finding that human scroll patterns show higher B and near-zero M compared with most AI agents, except occasional similarity from the ChatGPT Agent. A LightGBM model using only B and M achieved 73.4% accuracy in classifying single-page interactions, indicating scroll behaviour can serve as a predictive feature for distinguishing bots from humans in controlled settings.