SURVEY ON SEMI SUPERVISED CLASSIFICATION METHODS AND FEATURE SELECTION

Neethu Innocent, Mathew Kurian

Abstract: Data mining also called knowledge discovery is a process of analyzing data from several perspective and summarize it into useful information. It has tremendous application in the area of classification like pattern recognition, discovering several disease type, analysis of medical image, recognizing speech, for identifying biometric, drug discovery etc. This is a survey based on several semisupervised classification method used by classifiers , in this both labeled and unlabeled data can be used for classification purpose.It is less expensive than other classification methods . Different techniques surveyed in this paper are low density separation approach, transductive SVM, semi-supervised based logistic discriminate procedure, self training nearest neighbour rule using cut edges, self training nearest neighbour rule using cut edges. Along with classification methods a review about various feature selection methods is also mentioned in this paper. Feature selection is performed to reduce the dimension of large dataset. After reducing attribute the data is given for classification hence the accuracy and performance of classification system can be improved .Several feature selection method include consistency based feature selection, fuzzy entropy measure feature selection with similarity classifier, Signal to noise ratio,Positive approximation. So each method has several benefits

Keywords: Semisupervised classification, Transductive support vector machine, Feature selection, unlabeled samples

DOI: https://doi.org/10.15623/ijret.2013.0212097