Application of a Selective Desensitization Neural Network to Concept Drift Problems

DOI Web Site 3 References Open Access
  • Ichiba Tomoki
    Graduate School of Systems and Information Engineering, University of Tsukuba
  • Horie Kazumasa
    Center for Computational Sciences, University of Tsukuba
  • Someno Shoichi
    Graduate School of Systems and Information Engineering, University of Tsukuba
  • Aki Tomohiro
    Graduate School of Systems and Information Engineering, University of Tsukuba
  • Morita Masahiko
    Faculty of Engineering, Information and Systems, University of Tsukuba

Description

<p>A selective desensitization neural network (SDNN) has a high function-approximation ability, low hyperparameter dependence, and suitability for online incremental learning. These properties suggest that an SDNN can deal well with temporal changes in the characteristics of data, or concept drift, although this has not been verified. In this study, we conducted experiments on online learning using an artificial dataset generated using a time-varying function and a real-world dataset of a stock prices index, and evaluated the effectiveness of an SDNN to solve concept drift problems. The results show that using the SDNN exhibited superior performance over the existing methods for both datasets, suggesting that an SDNN is highly suitable for certain types of concept drift problems.</p>

Journal

References(3)*help

See more

Related Projects

See more

Details 詳細情報について

Report a problem

Back to top