Open Conference Systems, CLADAG2023

Font Size: 
Multiple imputation for clustering on incomplete data
Vincent Audigier, Ndèye Niang

Last modified: 2023-07-07


We present how MI can be considered for addressing missing values in the context of clustering. For achieving this goal, we present a novel imputation method entitled FCS-homo, as well as a pooling method for the set of partitions obtained from each imputed data set. The proposed methodology is evaluated using a simulation study in comparison with state of the arts methods. We start by treating the case where the observations are generated from a gaussian mixture model with missing at random values. The study is completed by experiments based on various real data sets where the distribution of the data is unknown. These first results tend to show that multiple imputation is a efficient method for handling missing data in clustering, especially when the data distribution is unknown.