Predicting Phases in Business Cycles Under Concept Drift
Ralf Klinkenberg · 2003
For many tasks where data is collected over an extended period of time, its underlying distribution is likely to change. A typical example is information filtering, i.e. the adaptive classification of documents with respect to a particular user interest. Both the interest of the user and the document content change over time. Machine learning approaches handling this type of concept drift have been shown to outperform more static approaches ignoring it in experiments with different types of simulated concept drifts on real-word text data. In this paper, these approaches to learning drifting concepts are applied to the problem of classifying phases in business cycles. Their performance is compared to the more static approaches on real-world data for this classification task, in order to evaluate whether this domain also exhibits concept drifts and whether the concept drift approaches also allow performance gains in this domain. While previous studies were based on simulated concept drift scenarios, the experiments in this domain are not based on any simulated drift, but on the real concept drift inherent to this real-world data. Hence this paper provides significant support for the applicability of the proposed machine learning approaches to handling concept drift in realworld problems.