Year: 2017 | Month: December | Volume 4 | Issue 2
Ensemble Classifier based on Optimized Feature Matrix for Healthcare Dataset
Pratibha Mishra* and Megha Kamble
DOI:10.5958/2454-9533.2017.00011.4
Abstract:
The mining of health care data is important aspect for the forecast of critical disease like cancer. In health care data mining various tools and techniques are available and applicable from machine learning. Machine Learning offers popular effective technique of classification for the purpose of mining voluminous dataset of health care. This paper implements the different traditional classifiers decision tree, k-nearest neighbor, support vector machine and modified ensemble classifier random forest for classification of health/diseased entities from the UCI data set for Cancer. The paper proposes the feature matrix extracted from majority voting ensemble classifier random forest mapped to SVM. Then it is implemented on three variations of cancer data set. Random forest shows the great results in terms of reduction in features, overfitting by averaging several tree, and also algorithm show less variance by using multiple tree, reduce the chance of stumbling across a classifier that doesn’t perform well because of the relationship between train and test data. In Random Forest, randomness is introduced by identifying the best split feature from a random subset of available features. These available important features further classified by powerful supervised machine learning algorithm named support vector machine. The main purpose of feature selection approach is to select a minimal and relevant feature subset for the given dataset and maintaining its original representation. This approach enhances the performance of SVM classifier and give rise to modified the majority vote ensemble classifier. The proposed hybrid mechanism of random forest feature matrix and SVM classification has shown 1.3% increment in accuracy for reduced cancer data set and this is verified from three reduced cancer data sets. This paper also demonstrates better accuracy of proposed ensemble classifier by comparative analysis with existing classifiers mechanism.
Print This Article Email This Article to Your Friend
