Fast, finite, accurate and optimal WASD neuronet versus slow, infinite, inaccurate and rough BP neuronet illustrated via russia population prediction

Abstract

Russia population problem attracts great concerns to the future trend of population and the development of the nation. Conventional researches on Russia population prediction are usually based on the standard cohort-component method. Such a method only allows for several factors (fertility, mortality and migration rates), and then leads to the lack of all-sidedness in the prediction results. With outstanding generalization ability, the feedforward neuronet is considered to be a more appropriate substitute. Besides, the back-propagation (BP) is of the most widely-used feedforward neuronet. As the conventional back-propagation neuronet has some inherent weaknesses, in this paper, two types of improved feedforward neuronet are constructed for the Russia population prediction. More specifically, a type of 3-layer power-activated neuronet (PAN) equipped with the BP algorithm (BP-PAN) and a type of 3-layer PAN equipped with the weights-and-structure-determination (WASD) algorithm (WASD-PAN) are built on the basis of 2013-year (from 1AD to 2013AD) historical population data for the Russia population prediction. By a lot of numerical experiments, the future declining trend of Russia population in the next decade is predicted with the highest possibility. In addition, via the Russia population prediction, the comparisons on the performance between the WASD neuronet and BP neuronet are conducted and summarized.