Métodos de machine learning con algoritmos de clúster no supervisados, una alternativa de segmentación de las pymes colombianas para plantear estrategias de acuerdo con sus condiciones económicas

Abstract

This research created a new grouping alternative using machine learning tools such as K-means and agglomerative clustering models, based on financial information from 2016 to 2019 of 10,001 Colombian SMEs. From these models twelve clusters originated that have 98.44% of the evaluated data and it was determined that the model that presented the best clustering result was the agglomerative model which generates the following main groups: a first group with negative margins and a debt exceeding 61%, a second group starting with a range between -10% to 40% of its margins and a debt below 60%, and a third group with positive margins and a debt between 11 and 80%. Finally, these groups create strategies according to the economic conditions of each of them.