Prediction of Heart Attacks using Data Mining Techniques

Data mining tools have proven to be highly effective in recognizing and predicting early signs of heart disease, especially in countries where doctors lack extensive cardiovascular knowledge. This study introduces a new comprehensive combined dataset consisting of Cleveland, Hungarian, Swiss, Long Beach VA, and Stalog (Heart) datasets. The combined dataset contains 11 unique attributes and 918 observations. A total of eight different algorithms were used to analyze this newly combined dataset, including Logistic Regression (LR), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), K-Nearest Neighbor, Gradient Boosting, AdaBoosting, and XGBoosting. Using the Cleveland data set alone, the best prediction result was obtained by the LR algorithm with an accuracy of 82.6%, while the RF algorithm provided the best prediction accuracy of 89.9% using the combined dataset. In addition, we have identified the five most important features of the combined dataset using a deep exploratory data analysis. As a result of refining our dataset to include only the most relevant features, the XGB algorithm provided the highest prediction accuracy of 89.1%. Even though this decreases accuracy by 0.8%, it reduces time and money spent on non-essential tests.