Using MM principles to deal with incomplete data in K-means clustering

TL;DR

This paper proposes a method based on Majorization-Minimization principles to effectively handle incomplete data in K-means clustering, restoring data symmetry to improve clustering performance.

Contribution

It introduces a novel MM-based approach for imputing missing data in K-means, enabling the algorithm to work effectively with incomplete datasets.

Findings

The proposed method improves clustering accuracy on standard datasets.

The algorithm effectively restores data symmetry with missing attributes.

Source code and pseudo-code are provided for reproducibility.

Abstract

Among many clustering algorithms, the K-means clustering algorithm is widely used because of its simple algorithm and fast convergence. However, this algorithm suffers from incomplete data, where some samples have missed some of their attributes. To solve this problem, we mainly apply MM principles to restore the symmetry of the data, so that K-means could work well. We give the pseudo-code of the algorithm and use the standard datasets for experimental verification. The source code for the experiments is publicly available in the following link: \url{https://github.com/AliBeikmohammadi/MM-Optimization/blob/main/mini-project/MM%20K-means.ipynb}.