ANN Optimization Experiments for Classification

ANN Optimization Experiments for Classification

Copyright: © 2014 |Pages: 13
DOI: 10.4018/978-1-4666-6146-2.ch014
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

This chapter shows the application of PSO and GA algorithms for training the neural network using two datasets: XOR and Iris. Then the performance of both algorithms are compared and presented by figures. In addition, some of the other optimization algorithms such as Gravitation Search Algorithm (GSA) and Ant Colony Optimization (ACO) are explained.
Chapter Preview
Top

2 Neural Network Structure For Pso-Based Nn And Ga-Based Nn

Song and Gu (2004) mentioned that PSO has potential performance for modification of weights of neural network and it can be an acceptable alternative for BP algorithm. In this chapter, the feedforward NN trained with PSO is compared with BP rained with GA. A 3-layer NN is applied for the classification task by GA and PSO as training algorithms. The architecture of networks has the input, hidden and output layer. In each classification task, depending on the problem the numebr of nodes in each layer are different. The number of selected features in each database determines the number of input nodes and the number of classes defines the number of nodes in the output layer. However, determining the number of rules in the hidden layer still has no standard rule or theory.

In this chapter, the feedforward network trained with the PSO has been implemented. Results show that the convergence rate is faster and classification result is better in the experiments. Even it avoids the problems of GA. Below the required steps for performing the experiment is provided as the guidance:

  • 1.

    Construct PSONN based on PSO program for Sombrero function optimization.

  • 2.

    Construct GANN.

  • 3.

    Comparison between PSONN and GANN.

  • 4.

    Evaluate performance of 2 and 3.

The experiment is performed in 2 stages: in the first stage the neural network is trained using PSO, and in the second stage GA is used for taring the neural network. The convergence time and the accuracy are calculated for comparing the performance of each approach in the classification (Nuzly, 2006).

Top

3 Experiments

Below the results of PSONN and GA results for classification of 2 databases are presented. 2 databases are XOR and Iris.

Complete Chapter List

Search this Book:
Reset