Comparative Study of Genetic and Memetic Algorithms for the Optimization of Customer Segmentation in Business Intelligence

Abstract

Customer segmentation is a fundamental component of Business Intelligence, enabling organizations to identify homogeneous customer groups for personalized marketing, customer relationship management, and strategic decision-making. However, the performance of conventional clustering algorithms is often constrained by their sensitivity to data characteristics and local optima, motivating the application of evolutionary optimization techniques. This study compared the effectiveness of Genetic Algorithm (GA) and Memetic Algorithm (MA) in optimizing selected clustering models for customer segmentation. Customer transaction data were gathered and preprocessed using the Recency–Frequency–Monetary (RFM) framework, after which K-Means, Hierarchical Clustering, and DBSCAN were implemented as baseline clustering models. Each baseline model was subsequently optimized using GA and MA, and their performances were evaluated using the Silhouette Score, Davies–Bouldin Index (DBI), Within-Cluster Sum of Squares (WCSS), and runtime. The results showed that evolutionary optimization substantially improved the performance of Hierarchical Clustering, increasing the Silhouette Score from 0.4219 to 0.4816, reducing the Davies–Bouldin Index from 0.8027 to 0.7324, and decreasing the WCSS from 4188.94 to 3803.29. In contrast, K-Means exhibited virtually identical performance before and after optimization (Silhouette Score: 0.4818 versus 0.4816), indicating that the baseline solution had already converged to a near-optimal clustering configuration. Although both evolutionary approaches produced comparable results, the Memetic Algorithm achieved the best clustering quality with the lowest DBI (0.7324) while maintaining similar computational performance to the Genetic Algorithm. DBSCAN-based optimization produced non-evaluable clustering outputs under the experimental conditions because valid cluster structures could not be consistently formed. The study was limited to RFM-based customer transaction data and three clustering algorithms. Future research should investigate larger and more diverse datasets, alternative evolutionary optimization strategies, and additional clustering techniques to further enhance customer segmentation in Business Intelligence applications. The novelty of this study lies in providing a unified comparative evaluation of Genetic and Memetic Algorithms across centroid-based, hierarchical, and density-based clustering models, thereby offering practical insights into the suitability of evolutionary optimization techniques for customer segmentation in Business Intelligence.