Optimal coding of high-cardinality categorical data in machine learning

Analyzing categorical data in machine learning generally requires a coding strategy. This problem is common to multivariate statistical techniques, and several approaches have been suggested in the literature. This article proposes a method for analyzing categorical variables with neural networks. Both a supervised and unsupervised approach were considered, in which the variables can have high cardinality. Some simulated data applications illustrate the interest in the proposal.