Short-term multidimensional forecasting in digital marketing

Abstract

The article presents the results of the construction of a method and model of short-term multidimensional forecasting based on a neural network, in which, due to the development of a learning procedure, in the process of learning a neural network, a training and target data matrix is formed, which provides the result of a forecast in real time.

The results of the analysis of research and publications determined the development of the modern economy both in the world and in Ukraine. The key directions of the development of the modern economy are identified, which include highlighting the issues of the effective functioning of enterprises in the conditions of the digital economy. It was determined that the solution to the problems of the application of intelligent information systems, the competitiveness of enterprises, the creation of analysis systems for the organization of monitoring of cloud platforms is connected with the application of digital marketing technologies.

The results of the research of methods and models of short-term forecasting became the basis for increasing the efficiency of subjects of economic activity.

The results of the application of technologies and models in the marketing analytical system for real-time operation are systematized and presented, which made it possible to build a structural model of the marketing analytical system. The proposed model uses databases where transactions are processed in real time. Data source systems provide information for further processing. Therefore, the proposed model provides real-time storage and processing of information with subsequent construction of multidimensional OLAP data cubes.

Based on the obtained data, a method and model of multidimensional forecasting was built, which includes three stages. At the first stage, analytical analysis of indicators and formation of a training matrix from selected values from historical OLAP slices. On the second, the matrix of target data is formed, based on values equal to the gradual increase of the time interval that determines the forecasting horizon. At the next stage, the learning process of the neural network takes place. The article presents the results of short-term multidimensional forecasting, which showed the accuracy of the obtained results.

The paper considers the solution to the problem of short-term multidimensional forecasting of a neural network as a tool for increasing the speed and processing of data while providing access to multidimensional data (OLAP). A neural network model is proposed for solving the problem of short-term multidimensional forecasting, which uses training and target matrices. Matrix data can be constructed of different dimensions from data that provide information from OLAP.