NEURAL NETWORK METHODS FOR ESTIMATING THE COSTS OF RESEARCH AND DESIGN AND SURVEY WORKS

Abstract

Summary. The article describes the neural network methods for estimating the costs of research and design and survey work in the construction of roads, allowing on the basis of data previously projected objects produce price-setting ranking factors according to their impact on the cost of research and development works undertaken for public-private partnership (PPP). Advantages of neural network methods is determined by the following circumstances : neural network models automatically take into account the mutual influence of the pricing factors ; Neural methods are completely free of subjective factors. Optimization of neural network allowed rank pricesetting parameters according to their impact on the cost of research , design and survey work under PPP . This causes the 4 "customer status", "Type of work", "Kind of competition" and "Road category" in the aggregate more than 87 % determined by the unit price of the project. Specific calculations show that the neural network allow very accurate ( with a relative error less than 0.2%) to describe most of the objects , and only a small fraction - less than 5 % of subsets , with significant error - from 9 % to 17 %.