VarSelLCM: an R/C++ package for variable selection in model-based clustering of mixed-data with missing values.

Area of Science:

• Machine Learning
• Statistical Modeling
• Bioinformatics

Background:

• Model-based clustering is crucial for data analysis, but selecting relevant features and the optimal number of clusters remains challenging.
• Existing methods often require pre-processing for mixed data types and missing values, limiting their applicability.

Purpose of the Study:

• To introduce VarSelLCM, a novel R package for comprehensive model selection in model-based clustering.
• To provide a robust solution for handling diverse data types and missing values within a unified clustering framework.

Main Methods:

• VarSelLCM employs classical information criteria for full model selection, encompassing feature selection and cluster number determination.
• The package intrinsically manages continuous, integer, and categorical data, including missing values under a Missing Completely At Random (MCAR) assumption.
• Mixture models are utilized for integrated data imputation and clustering.

Main Results:

• VarSelLCM enables simultaneous detection of relevant features and the optimal number of clusters.
• The method effectively handles datasets with mixed data types and missing values without pre-processing.
• The integrated data imputation capability enhances the robustness of clustering results.

Conclusions:

• VarSelLCM offers a powerful and flexible tool for advanced model-based clustering.
• Its ability to perform full model selection and manage complex data structures makes it valuable for various scientific applications.