Mixed Deep Gaussian Mixture Model: a clustering model for mixed datasets

Clustering mixed data presents numerous challenges inherent to the very heterogeneous nature of variables. A clustering algorithm should be able, despite this heterogeneity, extract discriminant pieces information from variables in order design groups. In work we introduce a multilayer architecture model-based method called Mixed Deep Gaussian Mixture Model that can viewed as an automatic way merge performed separately on continuous and non-continuous data. This is flexible adapted well or sense generalize Generalized Linear Latent Variable Models Models. We also new initialisation strategy driven selects best specification model optimal number clusters for given dataset. Besides, our provides low-dimensional representations which useful tool visualize datasets. Finally, validate performance approach comparing its results with state-of-the-art models over several commonly used