A Model for Time Series Clustering Optimization Based on Genetic Algorithm to Analyze the Behavior of Customers

Currently, organizations use data mining and business intelligence tools and techniques to analyze the behavior of their customers. Customer segmentation is a broad analytical tool used to identify distinct groups of customers. Because static segmentation leads to missing important patterns and trends in customer behavior over time. In this research, a model is presented that displays the behavior of each customer as a time sequence of the variables of purchase novelty, number of purchases, purchase amount, and customer cost. In fact, it also considers the time dimension of customer behavior. Then, using the genetic algorithm, optimal weights are found for each feature, and customers are segmented with clustering algorithms. The data used in this research is related to the transaction data of a payment service company during a period of thirty months. The results indicate that the best clustering result is achieved using spectral clustering algorithm by computing silhouette and Calinski-Harabasz metrics. These findings demonstrate that with optimal weighting, the genetic algorithm has been able to combine the features in a way that improves the silhouette metric to 0.91.