Bagging-AdaTSK: An Ensemble Fuzzy Classifier for High-Dimensional Data

Abstract

Using fuzzy systems to deal with high-dimensional data is still a challenging work, even though our recently proposed adaptive Takagi-Sugeno-Kang (AdaTSK) model equipped with Ada-softmin can be effectively employed to solve high-dimensional classification problems. Facing high-dimensional data, AdaTSK is prone to overfitting phenomenon, which results in poor performance. While ensemble learning is an effective technique to help the base learners to improve the final performance and avoid overfitting. Therefore, in this paper, we propose an ensemble fuzzy classifier integrating an improved bagging strategy and AdaTSK model to handle high-dimensional classification problems, which is named as Bagging-AdaTSK. At first, an improved bagging strategy is introduced and the original dataset is split into multiple subsets containing fewer samples and features. These subsets are overlapped with each other and can cover all the samples and features to guarantee the satisfactory accuracy. Then, on each subset, an AdaTSK model is trained as a base learner. Finally, these trained AdaTSK models are aggregated together to conduct the task, which results in so-called Bagging-AdaTSK. The experimental results on high-dimensional datasets demonstrate that Bagging-AdaTSK has competitive performance.