Diseño de un modelo para la predicción de ataques cardíacos mediante técnicas de aprendizaje automático

Abstract

A heart attack is ischemic necrosis of the heart, which is usually caused when the arteries supplying the heart are obstructed. Early detection of this cardiovascular disease increases the chances of saving the lives of many people and allows them to take measures to take care of their health. Over the years, with the help of technology, great advances have been made in the area of healthcare, such as Artificial Intelligence (AI), which has made it possible to improve patient care by streamlining processes and obtaining greater precision in medical diagnoses. X-rays, preparations with respect to pathological anatomy and medical examinations are being used to assist in the process of diagnosis and treatment of patients through the use of machine learning. Allowing in this way that there are projects dedicated to explore the applications of Artificial Intelligence in some healthcare areas such as, the care area, which is responsible for disease prevention, diagnosis, treatment and monitoring of patient health; the area of continuing education, which allows specialists to interact with virtual training and learning environments; the research area, which allows the search for evidence of the usefulness of AI in health through the use of medical information, helping to develop programs for numerous diseases, and finally the management area, in which large amounts of historical data are analyzed, which facilitates the treatment of both material and human resources in a better way, against a specific situation. This work is oriented within the field of data science, and consists of the development of a prediction model of heart attacks, from a dataset of clinical data collected from patients, which was obtained from the Kaggle site, which has 65535 records and is composed of 13 variables such as, age, gender, weight, height, body mass index, systolic blood pressure (high), diastolic blood pressure (low), cholesterol, glucose, smoking, drinking, exercise and cardio. And making use of machine learning techniques, since these techniques allow the identification of patterns in the data, which will allow the model to learn from the observations and most importantly to make predictions. Some Machine Learning algorithms were used such as; Logistic Regression, Random Forest, MultiLayer Perceptron, Extreme Gradient Boosting, Gaussian Naive Bayes, Support Vector Machine, Decision Tree, Gradient Boosting and Light Gradient Boosted Machine, to determine which one performs a better prediction, to choose the algorithm the performance metrics were used which were: Confusion matrix, accuracy, precision, sensitivity, F1 score and the area under the curve, giving us better results the Gradient boosting algorithm, having the following results: accuracy 72.33%, precision 74.10%, sensitivity 69.83%, F1 score 71.90% and area under the curve 72.36%. Which was saved in a .pkl extension file to be used in the creation of a dashboard on a web page, which allows the entry of data according to the variables of the dataset mentioned above, and predict whether a person may have a heart disease or not, and in turn taking into account the data of the variables to make recommendations to the patient.