A new improved clustering method of incomplete data based on K-nearest neighbors

Clustering algorithm has been widely studied and applied in machine learning. Most of the clustering algorithms are only suitable for complete data sets; however, real-world data sets may have missing values for a variety of reasons. To solve this problem, this article propose a fuzzy C-means (FCM) clustering algorithm based on K-nearest neighbors(KNN) imputation. Unlike the two-stage strategy that executes imputation and clustering separately, this article take missing values as the optimization variable of the objective function to make imputation and clustering negotiate with each other to achieve the optimal. The proposed algorithm has a good effect on clustering for incomplete data. The experimental results on UCI data sets can verify that the proposed approach is obviously superior to several other approaches.