Publication:
Multilayer perceptron neural network for triangular waveform classification

Loading...
Thumbnail Image
Date
2008-03-01
Authors
Teng, Ling Huey
Journal Title
Journal ISSN
Volume Title
Publisher
Research Projects
Organizational Units
Journal Issue
Abstract
Triangular waveform is applicable to many fields, such as clock signal synchronization. The clock signal is very important to synchronize the operations among the modules in the VLSI circuits. The smoothness of the triangular waveforms produces a lower frequency component which is able to translate into significantly lower power consumption and induced noise when compared with a square clock signal. The waveform data was obtained from the UCI Machine Learning Repository. They consist of 21 input attributes with continuous values. The output represents 3 types of triangular waveforms denoted as Class 0, Class 1, and Class 2. Each class is generated from a random convex combination of two of the three ‘base’ triangular waves as described by Breiman. This project presents the classification of triangular waveform using an Artificial Neural Network (ANN) model as an intelligent system. The Multilayer Perceptron (MLP) ANN is proposed to be used in developing the intelligent classification system. The MLP network is trained using two learning algorithms; Levenberg-Marquardt Back propagation (LM) and Resilient Back propagation (RP) algorithms. The proposed MLP Neural Network architecture is trained and tested using the MATLAB software package. In this study, the results show that both RP and LM algorithms produce the same percentage correct classification of 89.40%. But RP algorithm was chosen to be a better learning algorithm compared to the LM algorithm in performing the waveform classification because RP requires only 27 neurons in the hidden layer, meanwhile LM requires 40 hidden neurons. This study also demonstrates the superiority of an MLP network for the task as compared to the Optimal Bayes classification method which only obtains 86.00% of correct classification which using 300 data for the training.
Description
Keywords
Citation