Studying the Impact of Changing Consumer Behavior During Crisis Periods Through Store Classification

Abstract

Since customer behavior changes unpredictably during crisis periods such as pandemics, many sectors have been affected differently. The retail sector in particular has been one of the most affected sectors. Retail companies that could not determine the right strategies against customer behavior change were in a difficult situation, and some even had to close down. The inability of consumers to do physical shopping for reasons such as socializing, experiencing products and interacting during the pandemic process required an understanding of changing consumer needs. In this study, to determine the changes in customer purchasing behaviors during the pandemic period, using the sales data of a company operating in the women’s clothing sector and whose sales loss approached 50% during the pandemic period, separate stores were divided into clusters using machine learning methods for the pre-pandemic and pandemic period. The clusters formed were examined and the stores in different clusters were determined depending on customer purchasing behavior. The aim of the study is to ensure that the company segments its stores correctly to gain competitive advantage. Firms will be able to determine the right strategies against changing consumer behavior through a correct store segmentation. First, stores that do not belong to any classification group are clustered using unsupervised machine learning methods. No significant change was observed in the clusters formed before and during the pandemic. This indicated that the pandemic had a similar effect on all stores. Then, pre-pandemic, pandemic period and both periods data were analyzed using 7 different machine learning classification algorithms. The results obtained were compared. For all three analyses, the random forest algorithm gave the highest accuracy rate. The random forest algorithm with the highest accuracy was hybridized with 3 different classification algorithms. The hybrid model consisting of random forest and support vector machine gave the highest accuracy rate (90%) for the period including all data for store classification. Thanks to the hybrid model created with random forest and support vector machines, companies can be advantageous against other companies in the competitive environment by creating separate strategies for each store class.