Gender-Specific Predictive Modeling of Bidirectional Risk between Heart Disease and Diabetes Mellitus using Machine Learning

Cardiovascular diseases (CVD) and diabetes are two of the most challenging global health problems and often coexist to escalate morbidity and mortality. Early prediction of these diseases using machine learning makes way for early intervention itself. In this paper, the authors provide a comparative study of the Logistic Regression method and the Random Forest approach used to predict two of the most challenging intersections of clinical diseases: (1) the possibility of developing cardiovascular diseases when already suffering from diabetes and vice versa, and (2) the possibility of developing diabetes when already affected with cardiovascular diseases. In this work, the structured data has been used to segregate the community of people into four groups: Everyone, Pregnant Women, Women, and Men. The results indicate the possibility of large differences in probabilities of predicting diseases across the sexes and also across the group of pregnant and non- pregnant women and men. The results also support the fact that the Logistic Regression approach outperforms the Random Forest approach in predicting the possibility of developing cardiovascular diseases across almost all groups. This suggests the applicability of the Logistic Regression approach when the predicting factors are mostly linear in nature. However, the performance of the two models was found equally efficient in predicting the possibility of developing diabetes. This suggests that the effectiveness of the predicting model might depend upon the characteristics of the particular disease being studied.