Proactive-Reactive Prediction for Data Streams

Prediction in streaming data is an important activity in various branches of science such as sociology, economics and politics. Two major challenges offered by data streams are (1) the underlying concept of the data may change over time; and (2) the data may grow without limit so that it is difficult to retain a long history of raw data. Previous research has mainly focused on manipulating relatively recent data. The distinctive contribution of this paper is in three folds. First, it uses a measure of conceptual equivalence to organize the data history into a history of concepts. Transition patterns among concepts can be learned from this history to help prediction. Second, it carries out prediction at two levels, a general level of predicting each oncoming concept and a specific level of predicting each instance’s class. Third, it proposes a system RePro that incorporates reactive and proactive mechanisms to predict in streaming data with efficacy and efficiency. Experiments are conducted to compare RePro with representative existing prediction methods on various benchmark data sets that represent diversified scenarios of concept change. Empirical evidence offers inspiring insights and suggests the proposed methodology is an advisable solution to prediction for data streams.