Machine Learning Approaches for Early Prediction of Cardiac Diseases

Abstract

CVDs are an enormous murderer of individuals in all parts of the world, and thus the need of the successful early diagnosis and intervention measures. The weakness of the traditional risk prediction models is that they cannot handle high dimensional and multi-dimensional data which has the tendency of giving an inefficient prediction. The paper presents an argument on how machine learning (ML) algorithms can be used to determine the presence of CVDs at a young age with the presence of multimodal health data, i.e., the logistic regression, random forests, support vehicle machines (SVM), and the neural networks. Models were trained and evaluated by the UCI Heart Disease Dataset and Framingham Heart Study Dataset. Preprocessing involved the main procedures of missing data management, categorical variables data normalization and encryption. To select the best predictors, the feature selection was done using Recursive Feature Elimination (RFE) and mutual information. Comparison of the models has been done based on accuracy, precision, recall, F1-score and based on Area Under the Receiver Operating Characteristic Curve (ROC-AUC). The findings revealed the better performance of the neural network model than the other models with accuracy 86, precision of 0.82, recall 0.80 and F1-score 0.81 and ROC-AUC 0.87 showing that neural network model is more predictive of CVDs. Random forest had then 85 percent and the lowest of the performance in logistic regression. The results indicate that ML and neural networks in particular can make a substantial contribution to the early detection and prediction of cardiovascular diseases which may be implemented as a potential device of clinical decision making and patient care.