Algoritmos de aprendizaje no supervisado para la estimación de la resistencia a la insulina y el síndrome metabólico en el adulto mayor de la ciudad de Cuenca

Abstract

In this degree work, Insulin Resistance (IR) and Metabolic Syndrome (MS) are explored in the Cuenca‘s elderly population, from the point of view of unsupervised classification methods. In the case of IR, five estimation methods were analyzed using a K-means classification on a population of 119 people older than 65 years old who underwent a two-point Oral Glucose Tolerance Test (OGTT). The K-means algorithm with K = 2 and K = 3 was applied in onedimensional experiments for the Homa-IR, Quicki, Avignon, and Matsuda methods. The results obtained allowed the development of a platform to aid in the diagnosis of IR. These findings were object of two publications (IV Congress of Information and Communication Technologies TIC-EC 2017, and IEEE ETCM 2017: 2nd IEEE Ecuador Technical Chapters Meeting). For the study of MS using Kohonen’s SOM, two types of normalization (binary and by rank) were analyzed for the inputs of the neural network using a population of 387 elderly people. The results, using a pre-processing by ranges allow a better classification of the population in all cases. This study allowed to select the type of pre-processing for the diagnosis of MS in the elderly population of the city of Cuenca using SOM and was the object of a publication in the V Congress REDU 2017 and the II Congress I+D+Ingeniería. The future work is oriented to validate the results obtained in other elderly populations.