Mining the change of customer behavior in dynamic markets

Identification of changes in customer behavior is a major challenge that must be tackled in order to survive in a rapidly changing business environment. For example, because information technology has advanced and data‐storage costs have declined, for the purpose of serving customers, numerous enterprises have employed information systems and have directly logged customer behavior in databases. This trend has motivated the development of data mining applications. Fuzzy quantitative sequential pattern mining is a functional data mining technique that is used for discovering customer behavioral patterns over time and determining the quantities of goods or services they purchase. The example term used in shopping ⟨[(Beer, Low)(Milk, High)] (Cola, Middle)⟩ means that customers will first buy Beer and Milk in Low and High quantities, respectively, and then purchase Cola in Middle quantities on their next shopping trip, where Low, Middle, and High are predefined linguistic terms assigned by managers. A term such as this one provides managers with general and concise knowledge related to customer behavior and allows them to rapidly make decisions in response, especially in a competitive setting. However, literature searches indicate that no previous study has addressed the issue of changes in fuzzy quantitative sequential patterns. The aforementioned example pattern might have been available last year but might not be used this year, and it could have been substituted by ⟨(Beer, Middle) {(Cola, Low)(Milk, Low)}⟩. If this knowledge is not renewed, managers might develop inappropriate marketing plans for their products or services and use inventory strategies that are outdated with respect to time and quantities. To solve this problem, we propose a novel change‐mining model that can be used for detecting changes in fuzzy quantitative sequential patterns. We conducted experiments in which we used real‐world and synthetic datasets in order to evaluate the proposed model. When the pattern change was detected using the real‐world dataset, the results showed that the model reveals 3 considerations that can help managers with their handling of products’ marketing and production. When we studied the model’s scalability by using the synthetic dataset, the results showed that even though all run times increased when parameter values were decreased, the model remained scalable.