Machine Learning for Ranking Factors of Global and Regional Protest Destabilization with a Special Focus on Afrasian Instability Macrozone

Based on the experience of previous studies, the authors use machine learning methods at two levels for evaluating predictors of instability. First, they analyze the factors that lead to instability in general; second, they focus on the factors that influence the intensity of instability. Their analysis relies on data on mass protest destabilization. The system for assessing predictors of nonviolent destabilization is modernized and a two-level model is developed for ranking the factors of instability. After that, using Shapley vectors, all predictors within the final model are estimated and quantified. The authors analyze several subsamples: the world as a whole, the World System core and periphery, and the Afrasian instability macrozone. The result shows that the division of the original database into world-system zones, as well as specifying the Afrasian zone as a separate entity makes sense. The results obtained through machine learning are further cross-validated with more traditional regression models.