Predicción de riesgo cardio vascular a través de los algoritmos supervisados: KNN, Árboles de decisión y Random Forest.

Abstract

Abstract: This thesis is aimed at predicting cardiovascular risk through supervised Machine Learning algorithms, such as KNN, Decision Trees and Random Forest, models noted for their ability to identify complex patterns and make accurate predictions in data sets related to clinical variables and cardiovascular risk factors. In a context where cardiovascular diseases are the main cause of mortality in Ecuador, a database with relevant information on cardiovascular risk factors was used, and through a process that included exploratory analysis, data preprocessing, application of balancing techniques and refinement of hyperparameters by means of Grid Search, the proposed models were implemented and evaluated. The results indicate that the balancing techniques, especially the hybrid use of ROS and SMOTE, were important to improve the performance of the models, precisely in the minority classes, likewise, the elimination of highly correlated variables optimized the predictive capacity of the algorithms, resulting that in experiment three, the Random Forest algorithm achieved an accuracy of 0.83 and a better performance in the handling of unbalanced data. This study demonstrates the potential of supervised algorithms as effective tools for cardiovascular risk prediction, providing a basis for the development of technological solutions, such as web or mobile applications, that support clinical practice and contribute to health prevention strategies.