Comparative results of attribute reduction techniques for thai handwritten recognition with support vector machines

Abstract

© Springer International Publishing Switzerland 2016. Data reduction is an important step in machine learning and big data analysis. The handwritten recognition is a problem that uses a lot of data to get the good results. Thus, the attribute reduction can be applied to improve the accuracy of classification and reduces the learning time. In this paper, the attribute reduction techniques are studies. These techniques are applied to the Thai handwritten recognition problems. Support vector machines (SVMs) are used to verify the results of 4 attribute reduction techniques, i.e., principle component analysis (PCA), local discriminant analysis (LDA), locality preserving projection (LPP), and neighborhood preserving embedding (NPE). All of these 4 techniques will transform the original attributes to a new space with the different methods. The results show that LDA is a suitable data reduction technique for classifying the handwritten character with SVM. Only 10 % of features can give the accuracy about 47.68 % for 89 classes of the characters. This technique may give a better result when the suitable feature extraction techniques are applied.