L. N. Yasnitsky, F. M. Cherepanov
We describe and examine successful attempts to predict the results of Russian presidential elections of 2008 performed a year before they took place. The development of the existing model is developed on the basis of the examples from history of Russian elections rather than foreign ones and as for the exit of neural network - the percentage of "for" votes was taken. Personal qualities of the candidates are set as major input parameters. It is concluded that the neural network trained on the domestic experience performs forecasts essentially coinciding with the projections made by neural network, trained on the experience of the presidential elections in France and the USA.
When the adequacy is proved, study of the model is conducted, major patterns of subject area are withdrawn and correlation between age of the candidate and his rating is forecast. An example of working out recommendations on improvement a well-known politician’s rating based on virtual experiments on the model is given.
All models are based on neural networks of a perceptron type with sigmoidal activation functions.