Comparing and tuning machine learning algorithms to predict type 2 diabetes mellitus
Gabriel Aguilera‐Venegas, Amador López-Molina, Gemma Rojo‐Martínez, José Luis Galán–García
Abstract
The main goals of this work are to study and compare machine learning algorithms to predict the development of type 2 diabetes mellitus. Four classification algorithms have been considered, studying and comparing the accuracy of each one to predict the incidence of type 2 diabetes mellitus seven and a half years in advance. Specifically, the techniques studied are: Decision Tree, Random Forest, kNN (k-Nearest Neighbours) and Neural Networks. The study not only involves the comparison among these techniques, but also, the tuning of the hyperparameters of each algorithm. The algorithms have been implemented using the language R. The data base used has been obtained from the nation-wide cohort [email protected] study. This work includes the accuracy of each algorithm and therefore the best technique for this problem. The best hyperparameters for each algorithm will be also provided.