A new AI framework claims it can predict not just whether a public event will happen, but how big it will get.
Researchers built auto-ibDLM, a deep learning framework that models public events as dynamic interaction networks, tracking how participants connect and grow over time. The system first describes network structure using network-science metrics, then compresses those features into a compact representation through what the authors call an auto-learning layer. A GRU-based module takes that representation and forecasts future participant growth. The team tested the approach on 13 real-world public event datasets and two public dynamic network datasets, reporting more than 97% accuracy and results that beat existing state-of-the-art methods on both accuracy and generalization.
Most event-forecasting tools answer a yes-or-no question: will this happen. This one tries to answer how far it spreads, treating growth as a byproduct of who is interacting with whom rather than a fixed set of external signals. If it generalizes, that distinction matters for anyone planning resources or risk response around protests, viral moments, or emergencies where scale, not just occurrence, determines the response.
The paper is an arXiv preprint, not yet peer-reviewed, and its 97% accuracy figure comes from benchmarks the authors selected themselves. Plenty of forecasting models look strong in a curated test set and wobble the moment real, messy events don't follow the pattern.