You are on page 1of 2

International Journal of Wisdom Based Computing, Vol.

1 (2), August 2011

59

Classification task by using Matlab Neural Network Tool Box A Beginners View
V. Arulmozhi
Department of Computer Science Tiruppur Kumaran College for Women, Tiruppur, India arultkc@yahoo.co.in
Abstract Neural network tool box consists of tools for designing, implementing, visualizing and simulating neural networks. This paper presents a beginners point of view of using this tool box for performing classification task for iris and wine data sets. Keywords- Neural Networks, IRIS dataset, WINE dataset, classification

build up to or are themselves used in engineering, financial, and other practical applications. More about neural network tool box can be had from the Tutorial of Neural Network Tool Box Software from math works [1] and [2]. Before starting the work of neural networks various papers are reviewed for clarification of doubts and how to do the task [3], [4], [5], [6], [7-10], [11]. III. CLASSIFICATION USING NEURAL NETWORK TOOL BOX

I.

INTRODUCTION

Machine learning, a technique which makes use of various disciplines like cognitive science, computer science, pattern recognitions and statistics, can be used to classify and predict data. Neural networks are used for applications where formal analysis would be difficult or impossible such as pattern recognition and nonlinear identification and control. Tool box supports feed forward networks, radial basis networks, dynamic networks, selforganizing maps and other networks. It would be impossible to cover the total range of applications for which neural networks have provided outstanding solutions [1]. At most basic level what we needed are neural network tools to classify / predict the physical, chemical and biological properties of various types of datasets. This paper presents a note on neural network tool box. This paper organized as follows. Section 2 presents a review on neural networks Section 3 presents a view of neural network tool box with examples and Section 4 concludes the paper. II. NEURAL NETWORKS A VIEW

The neural network tool box consists of a set of functions and structures which easily handle neural networks; we can use GUI or command-line operations or simple coding by using m-file. Neural networks are very good at pattern recognition problems. A neural network with enough elements (called neurons) can classify any data with arbitrary accuracy. The procedure is, first the user has to decide the structure of the MLP network architecture such as the number of hidden layers and neurons (nodes) in each layer. The activation functions for each layer are also chosen at this stage, that is, they are assumed to be known. The unknown parameters to be estimated are the weights and biases. Many algorithms exist for determining the network parameters. In neural network literature the algorithms are called learning or teaching algorithms, in system identification they belong to parameter estimation algorithms. The most well-known are back-propagation and Levenberg-Marquardt algorithms. Back-propagation is a gradient based algorithm, which has many variants. Levenberg-Marquardt is usually more efficient, but needs more computer memory. So better to use multilayer layer preceptron algorithm. First define the structure of the network, in the network, the activation functions are chosen and the network parameters, weights and biases, are initialized. Secondly parameter selection like error goal, maximum number of epochs are defined, third use training algorithm to train the network, fourth test the result by simulating output with measured input data. Then it is compared with the measured outputs and final validation is carried out with independent data. The MATLAB commands used in the procedure are newff, train, sim. MATLAB provides high-level network

Neural networks are composed of simple elements operating in parallel. These elements are inspired by biological nervous systems. As in nature, the connections between elements largely determine the network function. We can train a neural network to perform a particular function by adjusting the values of the connections (weights) between elements. Typically, neural networks are adjusted, or trained, so that a particular input leads to a specific target output. Neural networks have been trained to perform complex functions in various fields, including pattern recognition, identification, classification, and speech, vision, and control systems. Neural networks can also be trained to solve problems that are difficult for conventional computers or human beings. The toolbox emphasizes the use of neural network paradigms that

International Journal of Wisdom Based Computing, Vol. 1 (2), August 2011 creation function like newlin (create a linear layer), newp (create a perceptron) or newff (create a feed forward back propagation). They are particularly well suited for complex decision boundary problems over many variables. Therefore neural networks are a good candidate for solving the wine and iris classification problem.The neighborhood attributes will act as inputs to a neural network, and the respective target for each will be a n-element class vector. The network will be designed by using the attributes of neighborhoods to train the network to produce the correct target classes. The work flow of any problem can consists of six primary steps they are collection of data, create the network, initialize the weights and biases, train the network, validate the network and use the network. First load the dataset e.g. load wine it automatically arrange the dataset attributes X and classes y. then use P=X; then find its size by using size(P) then give T=y; then find its size by giving size(T). We need not split the data set for training and testing separately it could be done be the neural net works. The data preparation task consists of three stages like 75% training, 15% validation and 15% testing data by using the coding part [trainV,valV,testV]=dividevec(P,T,0.10,0.10). here P is inputs (attributes) and T is targets (classes). Then design the network by using patternnet=10; then use net=patternnet; create a network function like newlin(create a linear layer) or newp(create a perceptron) or newff(create a feed forward back propagation network). Then it will display the sub object structures. But here using net=newff(mimmax(P), size(T,1), {logsig, logsig}, trainlm) which creates the perceptron layer with correct input range type. Suppose any error like vector needed use ind2vec . Then define learning parameter like net.trainParam.epochs=100; net.trainParam.show=10; net.trainParam.Ir=0.05; then train the network and it provides lot of choices in training functions(gradient decent, gradient decent mentum,Levenberg-Marquardt, etc), use any two types of training batch-training and on-line training. For batch training use net.trainFcn = trainb; for on-line training use trainc. use two useful parameters net.trainParam.epochs = 100 which maximum number of times the complete dataset used for training,; net.trainParam.show = 10 which is the time between status reports of the training functions To evaluate the performance of the method, we carried out the experiment on the two data sets iris and wine as listed below Table 1. Classification Accuracy (ACC) in % Data sets Iris wine IV. feature 4 13 Class 3 3 Instance / Training / Validation / Testing 150 / 60 / 15 / 75 178 / 90 / 10 / 78

60

CC % 93% 91%

CONCLUSION AND FUTURE WORK

This paper proposes a beginners view of using matlab neural network to perform classification and predicts the results of iris and wine datasets. Neural networks are effective and efficient tool that has applications in many areas and disciplines. V. ACKNOWLEDGEMENT

The author is thankful to Mother Teresa Womens University, Kodaikanal and the Chairman, Principal, and all the staff members of Department of Computer Applications, Tiruppur Kumaran College for women, Tiruppur for their kind co-opearation and valuable support. REFERENCES
[1] Mark Hudson Beale,Martin T. Hagan,Howard B. Demuth, Neural Network Toolbox 7 Users Guide [2] http://www.igi.tugraz.at/lehre/EW [3] Li, X. and J. Gao, Pad modeling by using artificial neural network," Progress In Electromagnetics Research, PIER 74, 167180, 2007. [4] Progress In Electromagnetics Research, PIER 94, 83-104, 2009, A New Classifier For Polarimetric Sar Images Y.-D. Zhang, L.N. Wu, and G. Wei [5] F. Yaghouby, 2A. Ayatollahi and 3R. Soleimani , Classification of Cardiac Abnormalities Using Reduced Features of Heart Rate Variability Signal World Applied Sciences Journal 6 (11): 15471554, 2009 [6] Miller, A., B. Blott and T. hames, 1992. Review of neural network applications in medical imaging and signal processing. Medical and Biological Engineering and Computing, 30 (5): 449464. [7] Clayton, R., A. Murray and R. Campbell, 1994. Recognition of ventricular fibrillation using neural networks. Medical and Biological Engineering and Computing, 32 (2): 217-220. [8] Minami, K., H. Nakajima and T. Toyoshima, 1999. Real-time discrimination of ventricular tachyarrhythmia with Fouriertransform neural network. Biomedical Engineering, IEEE Transactions, 46 (2): 179-185. [9] Acharya, R. et al., 2004. Classification of cardiac abnormalities using heart rate signals. Medical and Biological Engineering and Computing, 42 (3): 288-293. [10] Yang, T., B. Devine and P. Macfarlane, 1994. Artificial neural networks for the diagnosis of atrial fibrillation. Medical and Biological Engineering and Computing, 32 (6): 615-619. [11] Goh,A.T.C., Back-propagation neural networks for modeling complex systems, Artificial Intelligence in Engineering, vol.9, no.3, pp.143-151,1995.

You might also like