Methods of analyzing consumer behavior based on multi-source data

Abstract

PURPOSE: The aim of the article is to develop a system for analyzing processes and data from various data sources based on machine learning methods.

DESIGN/METHODOLOGY/APPROACH: For data analysis, models for forecasting the level of sales and the pipeline gathering operations for the preparation of features, data, model training and its verification were designed. The data analysis pipeline is built of stages related to the preparation of features and data preparation. The learning process requires a specific division of data. The data set has been divided into three subsets, i.e., the training and validation data used in the learning process and the test subset used to verify the quality of the model.

FINDINGS: The results of the conducted research show that the use of this type of analytical methods allows for the creation of new business processes, adaptation of services and goods to customer requirements, or the appropriate location of products on the retail space in order to optimize the time of shopping (especially taking into account the pandemic situation).

PRACTICAL IMPLICATIONS: The models presented in the article can be used by combining sales systems and behavioral data related to the movement of customers in the area of the sales space, where it is possible to build systems that allow optimization of orders, the way of arranging goods and other customer behavior patterns.

ORIGINALITY/VALUE: A novelty is the construction of a multi-source model for data analysis, where appropriate predictive models were built to predict the level of sales with the use of machine learning algorithms.