We aim to predict activities of political nature influencing or reflecting societal‐scale behavior and beliefs by applying learning algorithms to Twitter data. This study focuses on capturing domestic events in Egypt from November 2009 to November 2013. To this extent we study underlying communication patterns by evaluating content and metadata of 1.3 million tweets through computationally supported classification, without targeting specific keywords or users from the Twitter stream. Support Vector Machine (SVM) and Support Distribution Machine (SDM) classification algorithms are applied to detect and predict societal‐scale unrest. Latent Dirichlet Allocation (LDA) is used to create content‐based input patterns for the SVM while the SDM is used to classify sets of features created from meta‐data. The experiments reveal that user centric approaches based on meta‐data outperform methods employing content‐based input despite the use of well established natural language processing algorithms. The results show that distributions over user centric meta information provide an important signal when detecting and predicting events. Applying this approach can assist policymakers and stakeholders in their efforts toward proactive community management.