Comparing the performance of different learning neural network algorithms to predict distribution pattern of Bemisia tabaci in cucumber fields of Behbahan

Today, describing the distribution patterns of insects using interpolation and estimation methods in order to explore the possibility of proportional control where they has gained the attention of many researchers. This study was performed to evaluate the MLP neural network algorithms and interpolation of population estimates of B. tabaci in areas not sampled and mapped its distribution. Information density of this pest was obtained by sampling in cucumber field of Behbahan. For evaluating ability of different neural network algorithms we used, mean square error and coefficient of determination and to evaluat the network with optimal algorithm we utilized a comparisson of parameters such as mean, variance, statistical distribution and the determination coefficients of linear regression between predicted values by the neural network and actual values. Results showed optimum performance of neural network white Levenberg-Marquardt algorithms was in Learning rate 0.26, Momentum Factor 0.75 and 11 neuron in hidden layer and no significant difference between the values of statistical characteristics (mean, variance) and differences in the statistical distribution of predict and the actual pest density. In other words, an artificial neural network with Levenberg-Marquardt could well learn whitefly density data model. Map of interpolation showed that the pest had cumulative distribution and proved possibility of site-specific pest control on this field.