Short-term prediction of international behavior using a Holland classifier

This paper describes the use of a machine learning system called a Holland classifier to make short-term predictions of international events. The Holland classifier generates and applies if-then type rules based on observed data sampled by it. The system is empirically tested using the COPDAB daily events data for the U.S.-European interactions for the period 1948-1978. The model is used to predict discrete sets of events for 20 days following a randomly chosen date on the basis of the previous 40 days of events. Generally, a fully self-organizing Holland classifier is able to achieve about the same accuracy as a mathematically optimized estimator.