Feature Selection and Model Evaluation for Heart Disease Prediction Using Ensemble Methods

Cardiovascular disease prediction uses various data and prediction models to predict an individual’s risk of heart disease. It is essential to preventive medicine because it allows doctors to prevent and avoid heart-related problems. Early scientists used a variety of methods to make predictions. These include machine learning, deep learning, data mining, IoT and other methods. Data mining techniques can extract meaningful patterns from more extensive data sets. Association rule mining (ARM) is a data mining technique that generates significant association rules that clinicians can use to explain the relationship between various risk factors and cardiovascular disease. In this study, we employed an extension of ARM called Weighted ARM (WARM) with 70,000 observations of heart disease data made available through Kaggle. In this process, a weight is calculated for each feature using different methods such as correlation matrix, chi-square test, information gain, information gain ratio (IGR), and correlation. Then, other data analysis techniques are used, and the model’s performance is evaluated. The brute force method is used to find the best combination of features to predict heart disease. Using an ensemble voting method combining KNN, decision tree, and Naive Bayes, an accuracy of 78.42% is achieved for the selected set of attributes.