An efficient stacking-based ensemble technique for early heart attack prediction

In many parts of the world, heart disease is the leading cause of death. Preventing or effectively managing cardiac disease often depends on its early detection. There has been a significant uptick in research towards using machine learning to estimate the probability of cardiovascular disease. Using a variety of classification methods and stacking as ensemble techniques, this work investigates the problem of predicting cardiovascular illness. A total of 1025 patients are used in the analysis, and their clinical data is broken down into 14 different categories (e.g., age, sex, chest pain kind, blood pressure, cholesterol levels, and more). The initial step of the analysis is to preprocess the data by filling in missing values, standardizing the numbers, and encoding the categories. After that, the information is segmented into a training set and a test set for the purposes of model building and testing. Logistic Regression, Decision Tree, Random Forest, Extreme Gradient Boost, Naive Bayes, and K-Nearest Neighbors (KNN) are the six classification methods used in the research. Accuracy, precision, recall, and F1-score are only some of the measures used to assess the efficacy of various classification methods. The findings reveal that random forest and decision tree both yields a 92.68% accuracy, with extreme gradient boost coming in as a close second at 90.73%. In the second portion of the research, ensemble approaches, and more especially stacking, are used to boost the classification models’ accuracy. The goal of stacking, a method of ensemble machine learning, is to increase prediction precision by using numerous models in concert. By training a meta-classifier on the predictions of the base models, stacking combines the predictions of multiple base models. The results demonstrate that stacking considerably enhances the efficiency of the original classifiers. The stacked model outperforms each individual classifier, with an accuracy of 98.53%.