Evolving Efficient Clustering and Classification Patterns in Lymphography Data through Data Mining Techniques

Abstract

Data mining refers to the process of retrieving knowledge by discovering novel and relative patterns from large datasets. Clustering and Classification are two distinct phases in data mining that work to provide an established, proven structure from a voluminous collection of facts. A dominant area of modern-day research in the field of medical investigations includes disease prediction and malady categorization. In this paper, our focus is to analyze clusters of patient records obtained via unsupervised clustering techniques and compare the performance of classification algorithms on the clinical data. Feature selection is a supervised method that attempts to select a subset of the predictor features based on the information gain. The Lymphography dataset comprises of 18 predictor attributes and 148 instances with the class label having four distinct values. This paper highlights the accuracy of eight clustering algorithms in detecting clusters of patient records and predictor attributes and highlights the performance of sixteen classification algorithms on the Lymphography dataset that enables the classifier to accurately perform multi-class categorization of medical data. Our work asserts the fact that the Random Tree algorithm and the Quinlan’s C4.5 algorithm give 100 percent classification accuracy with all the predictor features and also with the feature subset selected by the Fisher Filtering feature selection algorithm.. It is also stated here that the Density Based Spatial Clustering of Applications with Noise (DBSCAN) clustering algorithm offers increased clustering accuracy in less computation time.