Projected Clustering Methods for Predicting Heart Disease

Abstract

Supervised and unsupervised learning techniques are increasingly applied to improve medical decision-making. Medical-recorded data also have accumulated large amount of information about patients and their medical conditions. Relationship and patterns within this data could provide new medical knowledge. Unfortunately, few methodologies have been developed and applied to discover this hidden knowledge. In this paper, we propose projected clustering method for generating clusters of similar bio-signal patterns from medical data to be analyzed and the various classification methods for reflecting information of heart signal on the classification/prediction model. The experiments show that the optimal cluster is constructed by applying PROCLUS algorithm and it has from 0.881 to 0.9 f1-value index of prediction under test data.