A system for forecasting the volume of sales of residential real estate based on machine learning

Abstract

Any construction project contains a cash flow model, the purpose of which is to assess the ability of an enterprise to generate cash in the required amounts and within the time required for planned costs, to calculate revenue, profit/loss. The income from the implementation of the construction project, as well as its profitability, directly depend on the volume of real estate sales.  The article describes a system for forecasting the volume of real estate sales by a construction company in the regional market. This system is built on a neural network basis using the domestic Loginom Community analytical platform. To train the system, three groups of factors were used that can be quantified from official sources: external macroeconomic factors determined at the federal level, external regional and retrospective data downloaded from the corporate database of a construction company and characterizing the dynamics of residential real estate sales.  The system has a modular structure. The modular structure gives the system a universal character by allocating independent modules in the structure, which allow taking into account regional, federal and corporate input factors. The system is trained and has good forecast properties. The average relative error of forecasting is 6.89%.