Prediction by Fuzzy Clustering and KNN on Validation Data With Parallel Ensemble of Interpretable TSK Fuzzy Classifiers

For many application scenarios where raw and even multidomain training data can be easily collected, and at the same time, validation data (as ground-truth data) are available, it becomes naturally desirable for us to perform an enhanced classification/prediction on only validation data with the appropriate leverage of training data. In this article, a novel ensemble framework EP-TSK-FK of Takagi–Sugeno–Kang (TSK) fuzzy subclassifiers, is proposed to achieve the following distinctive characteristics: 1) each interpretable TSK fuzzy subclassifier on each training subset can be quickly built in parallel such that its outputs provide the values of the corresponding augmented features of the original validation data space; 2) as a novel ensemble method of fuzzy subclassifiers, EP-TSK-FK trains all the interpretable TSK fuzzy subclassifiers only once and does not explicitly reuse them while predicting a testing sample, which thereby reduces the computational complexity of the ensemble process for prediction; 3) after running the proposed iterative fuzzy c-means clustering algorithm iterative fuzzy C-means clustering (IFCM) on the augmented validation data to obtain the representative centroids, the fast classification/prediction of EP-TSK-FK on the testing samples is realized by using the $k$ -nearest neighbor (KNN) method on the representative centroids with the original features; and 4) enhanced classification performance by the IFCM & KNN method is theoretically revealed, and the experimental results on the benchmarking datasets indicate the effectiveness of EP-TSK-FK and its parallel learning method in the sense of enhanced classification performance, running time, and interpretability.