Why a 12-year-old forecasting paper has stood the test of time
A 12-year-old paper on forecasting civil unrest using pre-neural network ML beats modern models for evaluation rigor.
Amazon Scholar Aravind Srinivasan and 29 co-authors won the 2025 KDD applied-data-science test-of-time award for their 2014 paper on EMBERS. The system used five models (Bayesian classification, logistic regression) to fuse open-source data—news, social media, satellite imagery—and successfully forecast civil unrest in 10 Latin American countries. Its lasting contribution is a robust framework for evaluating probabilistic forecasts of complex real-world events, which remains relevant despite advances in AI models.
Why It Matters
It provides a crucial blueprint for validating modern AI predictions in high-stakes domains like political risk and public safety.