An Unsupervised Feature Selection Method Based on Improved ReliefF and Bisecting K-means
Junyuan Yang, Xiaoxiang Wang, Dongyu Wang · 2018
Due to the fact that a large number of data are unlabeled with many high-dimensional in practical machine learning problems, this paper proposes an unsupervised feature selection model BK-IRfF based on Bisecting K-means and the improved ReliefF algorithm considering the redundancy information between features, combined with the idea of iteration. The model uses Bisecting K-means to label the unlabeled data, uses improved ReliefF algorithm for feature selection, and iterates the above process until the desired subset of features is output. Experiments show that this unsupervised feature selection method has good performance.