The Principle of Transformation between Efficiency and Effectiveness: Towards a Fair Evaluation of the Cost-Effectiveness of KDD Techniques

摘要

Most of the KDD literature focuses on analyzing the effectiveness of KDD techniques, in the sense e.g. of reducing the classification error rate in the case of classification tasks. Efficiency issues are usually considered of secondary importance, it considered at all. In contrast, we focus on the cost-effectiveness of KDD techniques, i.e. on the trade-off between effectiveness (reduction of error rate) and efficiency (reduction of processing time). In particular, we show that a gain in efficiency can be transformed into a gain in effectiveness, and this principle can be used to evaluate the cost-effectiveness of KDD systems in a fair manner. We discuss the application of this general principle to evaluate the cost-effectiveness of two general kinds of KDD techniques, namely classification algorithms and attribute selection algorithms.