The development of an automated pattern recognition based on neural network / Irni Hamiza Hamzah, Mohammad Nizam Ibrahim and Linda Mohd Kasim

Pattern recognition techniques are used to automatically classify a variety of physical objects or abstract patterns. The capability of powerful personal computers and affordable and high resolution sensors (i.e.: CCD cameras, microphones and scanners) have fostered the development of pattern recognition algorithms in new application domains (i.e.: fuzzy logic, neural network and genetic algorithm). Based on this idea, the objective of this project is to develop an automated pattern recognition system based on neural network to recognize the pattern of loaded data file. MATLAB Version 7.0.4 Release 14 has been used as a programming language to build the system. The performance of single neural network with two different types of architectures (i.e.: neural network with single output and multiple outputs) have been evaluated and compared. Two types of data (i.e.: iris data and cervical cancer data) have been used to test the performance of the proposed system. The selected neural network architecture is the Multilayer Perceptron (MLP) network, which is trained with three different types of learning algorithms, namely the Levenberg Marquardt (LM), Bayesian Regression (BR) and Gradient Descent (GDX). The results obtained showed that, for both iris and cervical cancer data, the MLP network trained using LM for single output produced the highest overall accuracy of 100% with the least value of hidden neurons and epochs. In conclusion, the MLP network trained using LM for single output produced the best performance compared to BR and GDX. This project has developed a user-friendly pattern recognition system by using MATLAB GUI. The system is capable of accepting any type of loaded data file and input parameters chosen by the user. The system also provides a user-friendly attributes to be used by unfamiliar MATLAB programming language user.