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

Abstract

Data mining is an emerging area in the field of computer science research that holds sway over diverse dimensions in health care exhibiting advancements in clinical decisionmaking, malady prediction, disease prognosis and the like. This research work places emphasis on three major phases of Data mining viz, Clustering, Outlier Detection and Classification. We have analyzed the Cardiotocography dataset from the UCI Irvine Machine Learning Repository comprising of 2126 Fetal Heart Rate (FHR) and Morphology Pattern (MP) records with 21 predictor attributes providing life saving information on the state of the fetus in the womb. We classify the records into three/ten target classes for the FHR/MP records respectively and our analysis reports 100% classification accuracy for Random Tree and Quinlan's C4.5 algorithm in the case of the FHR dataset and 99.86% accuracy for the MP dataset. However the presence of outliers in the data deviate the classification accuracy. Hence we detect the outliers and reveal the improved classification rate of the classification algorithms. This paper also highlights the possible clusters in the Cardiotocography dataset that unearth the significance of each attribute in deciding the cluster generation. This work is the first attempt to analyze, compare and declare the impact of data mining techniques by detecting outliers, executing classification algorithms and clustering the records of the Cardiotocography dataset. © EuroJournals Publishing, Inc. 2012.