A temporal classification based predictive model of recurring societal events

The ubiquitous nature of social media provides a range of ways for people to express their opinions and intentions against various societal issues much more conveniently, and provides a platform on which they can organise events such as protests. There is currently a lack of techniques that are able to predict these societal events in a timely manner. This paper proposes an innovative model for forecasting recurring societal events to foster new technical capability for the law enforcement agencies and other related public services. The sparse temporal event data is systematically investigated using a sliding window approach to formulate both a feature period and outcome period to capture the predictability of recurring events. The model with multiple classifiers was validated in two different measures to illustrate the effectiveness, including the state of the art metrics in the EMBERS system. The success of the model will help in research and development of the more challenging predictive models that utilising online open data sources to generate accurate predictions. The temporal-classification based model can also be applied to the prediction of events in other domains.