A Novel M-Cluster of Feature Selection Approach Based on Symmetrical Uncertainty for Increasing Classification Accuracy of Medical Datasets

Abstract

In recent days, due to the advancements in technology, a massive amount of data is generating in every area of study, including the medical field. This massive amount of data contains a large number of attributes and instances in it. It is not an easy task for classification and prediction from this high dimensional data. Because, all the attributes in the dataset can't give an impressive result in classification and prediction. Now, it is unavoidable to reduce the high dimensional data for better classification result, which is possible by feature selection and reduction techniques .In this research paper, a novel M-Cluster feature selection (Mcfs) based on Symmetrical Uncertainty (SU) Attribute Evaluator is proposed for improving the classification accuracy of medical datasets. The proposed approach divides the total feature space into 'M' clusters, each cluster has a finite set of attributes in it without any duplication. Feature subset formed by proposed technique is tested using Dermatology and Breast Cancer medical datasets, and compared with an existing filter-based feature selection techniques(Information Gain (IG), Chi-Squared (Chi), Gain Ratio Attribute Evaluator (GR), ReliefF (Rel) ).Experimental results displayed an improved performance with some of the clusters formed by proposed method than existing methods. For experimenting proposed technique, KNN-Lazy learner, Naive Bayes (NB) Classifier, J48-Rule based learner, JRip -Tree based learners are used.

Heart Disease Prediction Using Glowworm Swarm Optimization and Support Vector Machine Classifier

“…In this research proposal, Improved Phase Adjustment based Glowworm Swarm Optimization Algorithm (ISAGSO) is implemented for the selection of features [19]. A relatively novel algorithm is the Glowworm Swarm Optimization (GSO), an optimized and nature-inspired algorithm.…”

Untitled

“…Researchers have proposed a novel M-Cluster feature selection (Mcfs) based on Symmetrical Uncertainty (SU) Attribute Evaluator for improving the classification accuracy of medical datasets. Resultant feature subset has been tested on Dermatology and Breast Cancer medical datasets [2]. Data mining approach has been proposed by researchers for the classification of lung cancer subtypes [3].…”

Untitled

“…Authors of the research article [6] applied the SMOTE with ensembling approaches for increasing the prediction rate of kidney disease data. Filter based feature selection algorithms have been applied by the researcher for the classification of SONAR signal data [4,8].Symmetrical Uncertainty based feature selection method have been applied over some of the medical dataset to derive the best features before applying classification algorithms [9,10]. Feature selection based on correlation coefficient and Symmetrical Uncertainty is proposed by the researchers and applied over various dataset belongs to diverse areas [11].…”