• International Journal of 

     Soft Computing and Software Engineering [JSCSE]

    ISSN:  2251-7545

    Prefix DOI  :  10.7321/jscse

    URL: http://JSCSE.com

     

    A Peer-Reviewed Journal 


      JSCSE
     
  •  The International Journal of 

    Soft Computing and Software Engineering [JSCSE]

   
 

Publication Year: [ 2011 ] [ 2012 ] [ 2013 ] [ 2014 ] [ 2015 ] [ 2016 ] [ 2017 ]


Advance Search    
Table of Contents [Vol. 5, No.1, Jan]




Seyyed Reza Sharafinezhad, Mohammad Eshghi, Habib Alizadeh
Doi : 10.7321/jscse.v5.n1.1
Page : 1 - 12
Show Summary
Abstract . In this paper, a new and powerful method for Blind Source Separation (BSS) for single channel mixtures is presented. This method is based on non-Gaussian nonnegative matrix factorization (NG-NMF) in which modified Hilbert spectrum is employed. In the proposed algorithm, the Adaptive EEMD (AEEMD) is introduced to transfer the signal to the Enhancement Intrinsic Mode Functions (EIMF). The Hilbert spectrums of EIMFs are used as artificial observations. In order to make estimated spectrum of EIMF of sources using NMF, the maximization of Non-Gaussianity is used. Then, spectra of estimated oscillation modes are transferred to the time domain by the inverse Hilbert spectrum (IHS). In order to cluster of these oscillation modes, k-means clustering algorithm based on KLD (Kullback Leibler Divergence) is used. The simulation results indicate that the proposed algorithm performs the separation of speech and interfering sounds. from a single-channel mixture, successfully.
Keyword : Blind Source Separation (BSS) ; Non-Gaussian Nonnegative Matrix Factorization ; Adaptive Ensemble Empirical Mode Decomposition ; modified Hilbert spectrum (HS)




Leena Patil, Mohammed Atique
Doi : 10.7321/jscse.v5.n1.2
Page : 13 - 30
Show Summary
Abstract . Feature selection is a challenging problem in the field of machine learning, pattern recognition and data mining. Feature Subset Selection becomes an important preprocessing part in the area of data mining. In rough set theory, the problem of feature selection, called as attribute reduction, aims to retain the discriminatory power of original features. A large number of features is the problem in text categorization. Most of the features are noisy, redundant, relevant or irrelevant noise that can mislead the classifier and it may have different predictive power. Therefore, feature selection is often used in text categorization. It is most important to reduce dimensionality of the data to get smaller subset of features and relevant information within efficient computational time as time complexity is the major issue in feature selection. To deal with these problem many feature selection algorithms are available, still such algorithms are often computationally time consuming, and possess the problem of accuracy and stability. To overcome these problems we developed a framework based on neighborhood positive approximation rough set for feature subset selection in which the size of the neighborhood depends on the threshold value δ. In the proposed framework we obtain several representative and rank preservation of significance measures of attributes. In this paper firstly document preprocessing is performed. Secondly, a neighborhood positive approximation is used to accelerate the attribute reduction. Thirdly result validations based on classifiers are performed. Experimental results show that the improved feature selection based on neighborhood positive approximation rough set model becomes more efficient in terms of the stability, computational time and accuracy in dealing with large datasets.
Keyword : Introduction, document Preprocessing, Feature Sele