Better Learning of Supervised Neural Networks Based on Functional Graph – An Experimental Approach

Multilayered feed forward neural networks possess a number of properties which make them particularly suited to complex problems. Neural networks have been in use in numerous meteorological applications including weather forecasting. As Neural Networks are being more and more widely used in recent years, the need for their more formal definition becomes increasingly apparent. This paper presents a novel architecture of neural network models using the functional graph. The neural network creates a graph representation by dynamically allocating nodes to code local form attributes and establishing arcs to link them. The application of functional graph in the architecture of Electronic neural network and Opto-electronic neural network is detailed with experimental results. Learning is defined in terms of functional graph. The proposed architectures are applied in weather forecasting and X-OR problem. The weather forecasting has been carried out based on various factors consolidated from meteorological experts and documents. The inputs are temperature, air pressure, humidity, cloudiness, precipitation, wind direction, wind speed, etc., and outputs are heavy rain, moderate rain and no rain. The percentage of correctness of the weather forecasting of the conventional neural network models, functional graph based neural network models and the meteorological experts are compared.