Predição de doença cardíaca e identificação de fatores de risco com técnicas estatísticas e de Machine Learning

Abstract

Cardiovascular diseases (CVDs) are one of the leading causes of global mortality, representing a silent challenge to public health, including in Brazil. Given this scenario, this study’s objective is to develop a predictive model capable of forecasting the presence of heart disease and identifying the main associated risk factors, thereby contributing to prevention and risk management strategies. To achieve this, the methodology involved exploring data and fitting classification models using traditional statistical techniques and machine learning (Logistic Regression, Random Forest, and K-Nearest Neighbors - KNN). These methods were evaluated using metrics such as F1-score and AUC-ROC. For the identification of risk factors, Cramer’s V was employed to measure the association between predictor variables and the target variable. The results highlighted the predictive capability of logistic regression with variable selection, which showed the best balance between classes, with an F1-score of 0.23 for the positive class and an AUC-ROC of 0.7859. The Random Forest and KNN models, while efficient in the majority class, demonstrated inferior performance in detecting positive cases. The analysis of risk factors, in turn, reinforced the association of the disease with age, hypertension, diabetes, high cholesterol, stroke, and sleep problems. In conclusion, the study demonstrates the feasibility of applying machine learning methods in heart disease prediction, highlighting the importance of variable selection to improve the predictive capacity of models in scenarios of class imbalance.