Trends in Unsupervised Methodologies for Optimal K-Value Selection in Clustering Algorithms

Clustering algorithms are a powerful machine learning tool when working with large datasets, as they allow data to be grouped according to certain characteristics without the need to manually label the data. These algorithms generally request the number of clusters to be formed (k) as a parameter of the model and, while in some instances it is possible to indicate this number manually, most situations require this estimation to be an unsupervised task. The most widespread techniques offer acceptable results, but there is still much room for improvement. This study highlights their main shortcomings and reviews some of the advances in the estimation of this parameter presented in recent years, exploring their advantages and limitations.