Heart Disease Dataset Clusterization.

Area of Science:

• * Data Science
• * Machine Learning
• * Medical Informatics

Background:

• * Patient similarity is crucial for clinical decision-making.
• * Clustering methods offer potential for grouping similar patients.
• * Interpretability and method variability pose challenges in clinical applications.

Purpose of the Study:

• * To evaluate the utility of patient clustering for improving predictive modeling.
• * To compare different clustering algorithms (k-means, Agglomerative, Spectral) on a real-world dataset.
• * To assess the impact of cluster assignments as features in predictive models.

Main Methods:

• * Applied k-means, Agglomerative, and Spectral clustering to a dataset.
• * Evaluated cluster quality and correlation with data attributes.
• * Developed and compared predictive models using k-nearest neighbors (KNN) and Artificial Neural Network (ANN), incorporating cluster results versus original target attributes.

Main Results:

• * Clustering revealed patient subgroups correlating with specific data attributes.
• * Predictive models utilizing cluster assignments as features achieved higher F-scores compared to models using the original target attribute.
• * The chosen clustering methods produced distinct groupings, highlighting the impact of algorithm choice.

Conclusions:

• * Cluster analysis can enhance patient similarity identification.
• * Integrating clustering results into predictive models improves their performance.
• * Further research is needed to optimize interpretability and clinical integration of clustering outcomes.