Indian Journal of Science and Technology

Abstract

Background/Objectives: Medical Decision Support System (MDSS) is a diagnostic interface which provides computer assisted information retrieval as well as may support excellence decision making, to stay away from human error. Even if human decision-making is frequently most advantageous, but it is poor when there are vast amounts of data to be classified. Also capability and accuracy of decisions will decrease when humans are set into pressure and massive work. Forever there is a need and scope for a better MDSS. Methods/Statistical Analysis: Cluster analysis is a method of grouping of objects keen on different groups, has proved to be a valuable tool for identifying co-expressed genes, biologically related groupings of genes and patterns. K-means, Hierarchical and Fuzzy c-means are various clustering techniques have been employed to work as core part of MDSS. Findings: Proposed Weighted Class Based Clustering (WCBC) method is dependent on classifying properties of medical data itself. Weights are calculated on the basis of class value consequently increases separability by placing more number of instances of same class in same cluster. In this paper, the clustering algorithms K-means, Hierarchical, Fuzzy and Weighted Class based K-Means are examined for medical domains. Our finding is that on medical domains the Proposed Weighted Class Based Clustering outperforms others. Application/Improvements: The application of Proposed Weighted Class Based Clustering on medical datasets gave an insight into predictive ability of Machine Learning in medical diagnosis and there is a wide liberty that proposed approach can be used in RBF Neural Network for center calculation and data base Kernel Learning which is open area of research these days.

Keywords: Clustering, Fuzzy, Hierarchical, K-Means, MDSS, Weighted Class Based Clustering