Desarrollo de un modelo de clasificación automático en el software R para las etapas de la enfermedad renal crónica Mediante Machine Learning

Abstract

This project presents the development of an automatic classification model for the stages of chronic kidney disease using some machine learning techniques, such as support vector machines and neural networks. The UCI Machine Learning database, with approximately 400 instances of patients with and without CKD, was used to train and test the model. R software was used for modeling, but prior previously, the database was preprocessed in order to identify missing data and outliers in each of the variables. Similarly, data analysis and processing was performed in which new variables were added, such as the estimated glomerular filtration rate (eGFR), calculated with the CKD-EPI equation, to classify instances according to their value. Then, features were selected based on the level of correlation of the variables with eGFR, with the aim of improving the performance of the model, seeking to classify instances in the five stages of CKD. Thus, the best model obtained was the use of ANNs, with a total of 6 features, such as age, blood urea, serum creatinine, hemoglobin, packed cell volume and the number of red blood cells. The model generated an accuracy of 91% and a sensitivity of 93% for classifying test instances that did not have CKD. For individuals with stage 1 and 2, the model performed below 70%, making it not as reliable for these stages. In the case of stage 3, the model reached an accuracy of 76% and a sensitivity of 80%. On the other hand, in stage 4, it had an accuracy of 82% and a sensitivity of 95%. In addition, with an accuracy of 100% and a sensitivity of 57% for people in severe renal failure.