Hyperparameter optimization for cardiovascular disease data-driven prognostic system

Area of Science:

• Cardiovascular disease research
• Medical data mining
• Machine learning in healthcare

Background:

• Cardiovascular diseases (CVDs) are a leading cause of global mortality, necessitating improved prediction and diagnostic methods.
• Timely prognosis, considering patient history and lifestyle, is key for CVD prevention and management.
• Utilizing patient datasets for CVD risk factor analysis presents a significant challenge in modern medicine.

Purpose of the Study:

• To apply data mining and unsupervised machine learning techniques for analyzing Cardiovascular Disease Prognostic datasets.
• To evaluate the performance and classification accuracy of various machine learning models for CVD prediction.
• To determine the optimal number of clusters within CVD patient data using clustering methods.

Main Methods:

• Employed data mining via Orange software on a dataset of 918 adult patients (28-77 years old).
• Utilized supervised learning algorithms including k-nearest neighbors, support vector machine, random forest, artificial neural network (ANN), naïve bayes, logistic regression, stochastic gradient descent (SGD), and AdaBoost.
• Applied unsupervised clustering methods such as k-means, hierarchical, and density-based spatial clustering of applications with noise (DBSCAN) to identify data patterns.

Main Results:

• Stochastic Gradient Descent (SGD) and Artificial Neural Network (ANN) models demonstrated the highest performance, achieving a classification accuracy of 0.900.
• K-means and hierarchical clustering methods indicated that the Cardiovascular Disease Prognostic datasets could be effectively divided into two distinct clusters.
• The study highlights the strong correlation between model accuracy and the effectiveness of CVD risk prediction.

Conclusions:

• SGD and ANN are highly effective models for predicting cardiovascular disease risk with significant accuracy.
• The clustering of patient data into two groups suggests potential for distinct risk stratification.
• Accurate predictive models are essential for improving diagnostic capabilities and enabling timely preventive interventions for CVD.