Прогнозирование потоков экскурсионных групп музеев на основе модификации метода случайного леса

Логистика и управление цепями поставок. 2020. № 4 (99). С. 29–38.

Аслаханов А. Р., Pavlova E.

In recent years, both in Russia and in the world, there has been an annual increase in the number of museum visitors. The most popular exhibitions are visited by millions of people. In 2020, in the context of quarantine measures caused by the COVID-19 epidemic, the issue of managing the museum's visitors’ flows has become especially acute. If earlier the throughput of museums was limited by the maximum duration of a possible evacuation from the museums building, exhibitions space and the number of employees, who are working with the visitors, then in 2020, due to the observance of sanitary and epidemiological rules, the throughput of museums was further reduced. This determines the relevance of analytical solutions for museums since in order to manage visitor flows and adapt services to high demand, it is necessary to have an effective forecasting model that takes into account the determinism of demand by a number of factors.

The purpose of this paper is to develop a forecasting model for the number of excursion groups in specification museum-day-hour. A modification of random forest with the inclusion of more than 450 independent variables in the model is proposed as a forecasting method. The modification of the model consists in changing the mechanism for combining forecasts of trees in the forest in such a way that the weight of the tree in the model is inversely proportional to the measurement error of this tree. The proposed model is tested on the basis of data on more than 20,000 excursion groups of the State Russian Museum for the period 2018-2020. The proposed model showed high accuracy (36.6% WAPE and 0.5% BIAS). 

Keywords: музеи, машинное обучение, forecasting, прогнозирование, python, machine learning, service, сервис, museums, случайный лес, python, combining forecasts, комбинирование прогнозов, random forest