Performance Evaluation of Feed Forward Neural Network for Image Classification

Abstract

Artificial Neural Networks (ANNs) are one of the most comprehensive tools for classification. In this study, the performance of Feed Forward Neural Network (FFNN) with back-propagation algorithm is used to find out the appropriate activation function in the hidden layer using MATLAB 2013a. Random data has been generated and fetched to FFNN for testing the classification performance of this network. From the values of MSE, response graph and regression coefficients, it is clear that Tan sigmoid activation function is the best choice for the image classification. The FFNN with this activation function is better for any classification purpose of different applications such as aerospace, automotive, materials, manufacturing, petroleum, robotics, communication etc because to perform the classification the network designer have to choose an activation function.

Improving breast cancer prediction using a pattern recognition network with optimal feature subsets

“…MSE is the sum of squared distances between the observed and predicted values, which is the most commonly used loss function ( 27 ). Ullah et al ( 28 ) used the tan-sigmoid as the activation function. They also claimed that MSE might be the best parameter to find the best activation function ( 28 ).…”

“…Ullah et al ( 28 ) used the tan-sigmoid as the activation function. They also claimed that MSE might be the best parameter to find the best activation function ( 28 ).…”

Comparison of some artificial neural networks for graduate students

“…It is one of the widely spread networks that was called the front of the algorithm for the process of the algorithm in one direction only (from input to output) so that the weights of the subsequent units are not updated, the number of layers in it is not restricted [11] Let's assume we have the input the variables (X1, X2,X3, X4). ci = fi(wixi+bi) i=1,2,3,.... .....( 1…”

“…( 9) by taking partial derivation to update the weights as in Equation No ( . 5 ) When the least achievable error is reached, weights are adopted as parameters to the next units, and so on [11,16]. The number of hidden nodes is not restricted here and depends on the quality of the calculated processing.…”

Color space influence on ANN skin lesion classification using statistics texture feature

“…The information is presented as activation values, where each node is assigned a number, such that the higher the number, the higher the activation is. In a case of a feedforward network, the information is then transmitted forward throughout the network [11,12]. The activation value is transmitted from node to node but weighted in a certain manner.…”