An improved wavelet neural network for classification and function approximation
Loading...
Date
2011-01
Authors
Pauline, Ong
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Properly designing a wavelet neural network (WNN) is crucial for achieving the
optimal generalization performance. In this thesis, two different approaches were
proposed for improving the predictive capability of WNNs. First, the types of activation
functions used in the hidden layer of the WNN were varied. Second, the proposed
enhanced fuzzy c-means clustering algorithm-specifically, the modified point
symmetry-based fuzzy c-means (MPSDFCM) algorithm-was employed in selecting
the locations of the translation vectors of the WNN. The modified WNN was then
applied in the areas of classification and function approximation. In the context of
classification, the modified WNN was implemented to heterogeneous cancer and
diabetes classification using five different microarray datasets. The comparative
experimental results showed that the proposed methodology achieved an almost 100%
classification accuracy in multiclass prediction, leading to superior performance with
respect to other clustering algorithms. Performance comparisons with other classifiers
were made. An assessment analysis showed that this proposed approach outperformed
most of the other classifiers. In the context of function approximation, the modified
WNN was applied in approximating five different functions. Performance comparisons
indicated significant improvement in the approximation accuracy was accomplished by
the proposed WNN. Subsequently, performance comparisons with other methods in
approximating the same benchmark piecewise function were made. Evaluation
demonstrated the superiority of the proposed approach when compared with other
methods. A study of the proposed WNN in a real-world application, i.e. prediction of
blood glucose level for diabetics was also investigated.
A novel hybrid algorithm for edge detection was presented in this thesis. The
resulting algorithm, namely, wavClust, was then applied in the microarray image spot
segmentation. Comparisons with the classical spot segmentation methods were made.
Assessment analysis showed that the proposed wavClust algorithm was able to segment
all the donut-shaped spot, irregular spot and spots with intensity variations and different
noise types accurately, compared to the classical methods.
Description
Keywords
Wavelet neural , Network for classification , Function approximation