Abstract—Uncertainties in climate change have significant impacts on change in global temperatures having radiative forcing. The aerosols, emissions and concentrations from Green House Gases are the key drivers for such variations in the global climate. Carbon-dioxide is the most important anthropogenic Green House Gas having grown rapidly in the past three decades decreases the global energy. Carbon-dioxide is the most important anthropogenic Green House Gas having grown rapidly in the past three decades decreases the global energy. This paper presents an approach to determine the best performing Artificial Neural Network (ANN) algorithm for the forecast of carbon emissions from fossil fuels and flares and projects the emission rate. Various experiments are conducted to justify the swift, high performing, accurate and adaptive network amongst Simple Feedforward(FF), Multilayer Perceptron (MLP) with Backpropagation and Kohonen’s Self- Organizing Maps(SOM) methods . The stability, convergence and prediction accuracy of all the above methods are analyzed with different learning rule, transfer function and weigh update methods and tested statistically. It is observed that among the three network models, the MLP with Gradient Descent Backpropagation learning method and Tangent hyperbolic transfer function has stabilized in lesser number of epochs with higher prediction rate. The experiment results depict the future impact of emissions and vulnerability in climate change.

Index Terms—Artificial Neural Network (ANN), Feedforward (FF), Green House Gases (GHG), Kohonen Self- Organizing Map (SOM), Multilayer Perceptron (MLP).

Cite: Poornashankar and Vrushsen P. Pawar, "Assessment of Optimum Neural Network Architecture in Forecasting and Mining Carbon Emissions," International Journal of Computer and Electrical Engineering vol. 3, no. 2, pp. 219-225, 2011.