Self-organizing maps for exploration of partially observed data and imputation of missing values

The self-organizing map is an unsupervised neural network which is widely used for data visualization and clustering in the field of chemometrics. The classical Kohonen algorithm that computes self-organizing maps is suitable only for complete data without any missing values. However, in many applications, partially observed data are the norm. In this paper, we propose an extension of self-organizing maps to incomplete data via a new criterion that also defines estimators of the missing values. In addition, an adaptation of the Kohonen algorithm, named missSOM, is provided to compute these self-organizing maps and impute missing values. An efficient implementation is provided. Numerical experiments on simulated data and a chemical dataset illustrate the short computing time of missSOM and assess its performance regarding various criteria and in comparison to the state of the art.

Automated summary

In this paper, an extension of self-organizing maps to incomplete data is proposed, along with the estimation of missing values. The missSOM algorithm, an adaptation of the Kohonen algorithm, is introduced for computing these self-organizing maps and imputing missing values. Numerical experiments demonstrate the efficiency and performance of missSOM compared to the state of the art.