Cardiovascular Disease Analysis Using Different Machine Learning Techniques

Over the years, researchers have developed number of expert systems that helped cardiologists to diagnose patients more accurately by anticipating cardiac illness. In this research, it has been proposed, how building diagnostic model could use various machine learning techniques combined with hyperparameter optuna techniques to better predict cardiac disorders. In the current research we used enhanced Extreme Gradient Boosting (XGBoost) classifier, Random Forest (RF) and Support Vector Machine (SVM) Models, to detect cardiac diseases with basic key vital data points available. For any classifier model to be successfully applied, proper hyperparameter tuning is necessary. Hence, in order to maximize the accuracy, optuna optimization was employed which is highly effective technique for optimizing hyperparameters, for XGBoost, Support Vector Machines (SVM), and Random Forest methods. Optuna optimization is an effective technique for hyper-parameter optimization which is applied to optimize the XGBoost hyper-parameters. In order to verify the approach accuracy, we conducted a comparison with various other machine learning techniques using Optuna hyperparameter tuning. For building the proposed model, dataset from Kaggle was used comprising seventy thousand data instances. After the machine learning model was built using eighty to twenty split ratio between training data and test data, XGBoost model accuracy showed approximately eighty seven percent accuracy without Optuna optimization, while increase of one percent was conceded after applying Optuna optimization on the same dataset. Similarly, Random Forest and SVM models conceded eighty seven percent without Optuna and witnessed increase by 0.5 to 1 percent after optimizations. Area Under Curve (AUC) of the desired models showed up accuracy of 0.958 for all the three models including SVM, XGBoost and Random Forest. Based on the current research, it was concluded that the accuracy of the XGBoost with Optuna conceded better accuracy as compared to all techniques used in the previous researches. The overall accuracy achieved was little more than eighty eight percent. The study concluded that by using various data pre-processing techniques including heatmaps for data visualization, clustering with K-Modes and encoding helped in elimination of outliers which further resulted improved accuracy.