Comparison of linear Regression Methods and Artificial Neural Network Modeling of Population Dynamics Sunn Pest, Eurygaster integriceps Put. in Chadegan
عنوان دوره: دومین کنگره بین المللی حشره شناسی ایران
نویسندگان
چکیده
Artificial Neural Networks (ANN) as one of the modern modeling methods in recent years has received considerable attention. With the creation of powerful statistical techniques and neural networks, predictive models of distribution of ecology has developed rapidly. Neural networks are universal and flexible models for linear and non-linear systems. An important feature of neural models is that their implementation is not precluded by the theoretical distribution shape of the data used. Frequently, the performance of ANNs over linear or non-linear regression-based statistical methods is deemed to be significantly superior if suitable sample sizes are provided, especially in multidimensional and non-linear processes. This study aimed to investigate the relation between the Sunn pest Eurygaster integriceps occurrence and the meteorological factors and and to evaluate the effectiveness and performance of BP ANN (feed-forward backpropagation artificial neural network) and conventional models.Inputs ANN model were including mean daily temperature, mean daily relative humidity, average daily rainfall, altitude, date of sampling, wind speed, wind direction. Multilayer Perceptron network was used with back propagation algorithm and Levenberg Markvart learning techniques. The collected data randomly divided in three categories of training (70%), validation (15%) and testing (15%) and they used for train and test of two artificial neural networks, multi layer perception using back- propagation algorithm (MLP/BP) and nonlinear regression model. Various structures of neural networks by changing the input layer (7 models), was created. After sensivity analysis of seven inputs 4 inputs as input were elected. Sensitivity analysis indicated that the learning rate influenced the simulation performance of linear neural network model. The result show Among all inputs, average relative humidity the most important role. For comparison all models of the coefficient of determination (R2) and mean square error (MSE)were used. Accordingly, R2 and MSE for the multilayer perceptron model were calculated as 0.91 and 0.0076, respectively while these performance parameters for the linear regression was computed as 0.66 and 0.07, respectively.
کلیدواژه ها