A0526
Title: Automatic predictor of diabetes mellitus, type i and gestational, using machine learning techniques
Authors: Ricardo Stalin Borja-Robalino - University of Barcelona (Spain)
Antonio Monleon-Getino - Fundacio Bosh Gimpera (Spain) [presenting]
Karina Gibert - UPC (Spain)
Gladys Robalino-Izurieta - Universidad de Cuenca (Ecuador)
Brigith Borja-Robalino - Universidad de Cuenca (Ecuador)
Jorge Buelvas-Muza - Hospital Homero Castanier Crespo (Ecuador)
Raul Lopez-Torres - Universidad de Vic (Spain)
Carmen Serrano-Munuera - FUNDACIÓ HOSPITAL SANT JOAN DE DEU DE MARTORELL (Spain)
Abstract: Currently, data mining presents a massive development and optimization of Data Mining devices and algorithms that identify complex patterns through the development of systems that learn autonomously. On the other hand, the incidence and prevalence of diabetes have increased in recent decades in all countries, as a consequence of the decrease in life expectancy and the increase in unhealthy habits. In Ecuador, diabetes has a prevalence of 4.7\% in the population aged 10 to 59 years, while in Spain national surveys reflect a rate of about 8 out of 100 people. This research develops an automatic predictor of Diabetes, through Machine Learning techniques; becoming the first automatic predictive model at the national level for the prevention of Diabetes Mellitus, Type I and Gestational, trained with data from Hospitals in Ecuador and Spain. The choice of the model was based on the comparison of various techniques such as Logistic Regression, Linear Discriminant Analysis, Decision Trees, Naive Bayes, Support Vector Machine and Extreme Gradient Augmentation. The last model is found to be the most effective and efficient (83\% accuracy) for the prediction at the level of Ecuador and Spain. The evaluation of models used the ``EvaluaClas'' library for R, published previously and that allows the standardization of performance metrics for machine and deep learning classifiers.
