Model-based clustering with mixed/missing data using the new software MixtComp

Classification à base de modèles pour données mixtes et manquantes avec le nouveau logiciel MixtComp

Christophe Biernacki (1, 2) , Thibault Deregnaucourt (2) , Vincent Kubicki (2)

1 LPP - Laboratoire Paul Painlevé - UMR 8524
2 MODAL - MOdel for Data Analysis and Learning

The ``Big Data'' paradigm involves large and complex data sets where the clustering task plays a central role for data exploration. For this purpose, model-based clustering has demonstrated many theoretical and practical successes in a various number of fields. MixtComp is a new software, written in C++, implementing model-based clustering for multivariate missing/binned/mixed data under the conditional independence assumption. Current implemented mixed data are continuous (Gaussian), categorical (multinomial), integer (Poisson) and ordinal (specific model) ones. However, architecture of MixtComp is designed for incremental insertion of new kinds of data (ranks, functional, directional...) and related models. Model estimation is performed by a Stochastic EM algorithm (SEM) and several classical model selection criteria are available (BIC, ICL). Currently, MixtComp is not freely provided as an R package but is freely available through a specific user-friendly web interface (https://modal-research.lille.inria.fr/BigStat/) and its output corresponds to an R object directly usable in the R environment. Beyond its clustering task, it also allows us to perform imputation of missing/binned data (with associated confidence intervals) by using the mixture model ability for density estimation as well.

Domaines: Méthodologie [stat.ME]

Christophe Biernacki, Thibault Deregnaucourt, Vincent Kubicki. Model-based clustering with mixed/missing data using the new software MixtComp. CMStatistics 2015 (ERCIM 2015), Dec 2015, London, United Kingdom. ⟨hal-01249829⟩