Performance Analysis of Different Feed Forward Networks in Non-Linear Classification
Poornashankar
Poornashankar, MCA department, Indira College of Engineering & Management, Pune, India.
Manuscript received on April 06, 2013. | Revised Manuscript received on April 29, 2013. | Manuscript published on May 05, 2013. | PP: 332-337 | Volume-3, Issue-2, May 2013. | Retrieval Number: B1532053213/2013©BEIESP
Open Access | Ethics and Policies | Cite
© The Authors. Published By: Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Artificial Neural Networks (ANN) are recognized extensively as a powerful tool for most of the research applications including classification of heterogeneous data using function approximators. Identifying better neural classifier architecturefor a given input data depends on many factors, including the complexity of theproblem, the training set, the number of weightsand biases in the network and the error goal. Feedforward networks frequently exercise classification techniques for complex non-linear data. This paper presents a comparative study of different type of Feedforward neural networks such as Simple Feedforward networks, Pattern recognition networks and Cascade forward networks in classifying the global carbon emissions data. In this study the percapita carbon emissions of several countries are classified into low, medium and high category. Levenberg-Marquardt learning algorithm is used to train these networks as it is the fastest and first choice supervised learning algorithm with less training errors. Hyperbolic tangent activation function is used in this study because of their massive interconnectivity and enhanced processing performance. Experimental results show that simple Feedforward network outperformed in less number of epochs with higher classification accuracy.
Keywords: Green House Gases (GHG), Feed Forward network, Pattern Recognition Network, Cascade forward network.