| المستخلص: |
Neural networks have been one of the fascinating machine learning models, not only because of the fancy back propagation algorithm but also because of their complexity. Neural networks have not been popular, partly because they were, and still are in some cases, computationally expensive and partly because they did not seem to yield better results when compared with simpler methods such as support vector machines. Nevertheless, the neural networks have raised attention and become popular. In this paper, we are going to fit and classify neural networks using R package “neuralnet”, and fit a linear model "VGLM" and non-linear model “VGAM” models as a comparison applied on the environmental data. These data were collected from the Hunua Ranges, this is a small forest in southern Auckland, New Zealand. At 392 sites in the forest, the presence/absence of 17 plant species was recorded, as well as the altitude. Each site was of area size about 200 m2. We will study the effect of the presence/absence these species on the sea level for all areas in this forest. One of the most important procedures when forming a neural networks is data normalization. This involves adjusting the data to a common scale so as to accurately compare predicted and actual values. We used normalization process to scale data. Our neural networks has been created using the training data, then compare this to the test data to gauge the accuracy of the neural network forecast. We are used the neural networks and compares its results to tradition methods for regression and classification. Traditionally, the average MSE for the neural networks is lower than a linear model. But in our case the opposite is happened. Hidden layers use back propagation to optimize the weights of the input variables in order to improve the predictive power of the model. We study the effect of changing the number of hidden layers on the accuracy of our model. Also, traditionally an increasing of number of the hidden layers increase the accuracy of our model. Finally, A confusion matrix is used to determine the number of true and false generated by our predictions. This due to the accuracy rate of classification by divided the number of true on the total objects of a test data. A high accuracy rate expresses a good classifying.
|
