SciELO - Scientific Electronic Library Online

 issue31Performance Evaluation of a Self-Similar Model for Traffic on IEEE 802.11 NetworksInfluence of Subgrade and Unbound Granular Layers Stiffness on Fatigue Life of Hot Mix Asphalts - HMA author indexsubject indexarticles search
Home Pagealphabetic serial listing  

Services on Demand



Related links

  • On index processCited by Google
  • Have no similar articlesSimilars in SciELO
  • On index processSimilars in Google



Print version ISSN 0123-7799
On-line version ISSN 2256-5337


LOPEZ-SARMIENTO, Danilo A.; MANTA-CARO, Héctor C.  and  VERA-PARRA, Nelson E.. Least Square Support Vector Machine Classifier vs a Logistic Regression Classifier on the Recognition of Numeric Digits. TecnoL. [online]. 2013, n.31, pp.37-51. ISSN 0123-7799.

In this paper is compared the performance of a multi-class least squares support vector machine (LSSVM mc) versus a multi-class logistic regression classifier to problem of recognizing the numeric digits (0-9) handwritten. To develop the comparison was used a data set consisting of 5000 images of handwritten numeric digits (500 images for each number from 0-9), each image of 20 x 20 pixels. The inputs to each of the systems were vectors of 400 dimensions corresponding to each image (not done feature extraction). Both classifiers used OneVsAll strategy to enable multi-classification and a random cross-validation function for the process of minimizing the cost function. The metrics of comparison were precision and training time under the same computational conditions. Both techniques evaluated showed a precision above 95 %, with LS-SVM slightly more accurate. However the computational cost if we found a marked difference: LS-SVM training requires time 16.42 % less than that required by the logistic regression model based on the same low computational conditions.

Keywords : Support vector machine; least square; logistic regression; classifier; numeric digits.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )


Creative Commons License All the contents of this journal, except where otherwise noted, is licensed under a Creative Commons Attribution License