The effeciency of data types for classification performance of machine learning techniques for screening β-Thalassemia

Abstract

Performance of classification methods using Machine Learning Techniques majority depends on the quality of data were used in learning. The transformation techniques are used to increase the efficiency of classification because each type of data is suitable for different classification techniques. This study is aimed at providing comparative performance of different classification techniques by changing the type of data to find the appropriate type of data for each technique. The β-Thalassemia data is used for classifying genotypes of β-Thalassemia patients. The results of this study show that the types of data are Nominal scale which can be used as well for Bayesian Networks (BNs) and Multinomial Logistic Regression with the percentage of accuracy 85.83 and 84.25 respectively. Moreover, the data types which such as Interval scale can be used appropriately for K-Nearest Neighbors (KNN), Multi-Layer Perceptron (MLP) and NaiveBayes with the percentage of accuracy 88.98, 87.40 and 84.25 respectively. In the future, we will study the impacts of data separation to be used for classifying genotypes of patients with Thalassemia using the other classification techniques. ©2010 IEEE.