Innovative Customer Segmentation Based on Multi-Step Sequential Deep Clustering in the Telecommunication Industry

Nowadays, the exponential growth of the volume of data in telecommunication companies, causes the development of new techniques to discover information from existing datasets and implement data-driven strategies to increase loyalty and customer lifetime value. Due to the wide variety of customers, the marketing plans are unsuitable for all customer groups. Thus, customer clustering is used to divide customers into smaller homogeneous groups so that the marketing strategy can provide loyalty programs for each cluster in the form of exceptional services and rewards tailored to their behavior. This research aims to investigate the impact of dimensionality reduction algorithms on customer clustering performance in the telecommunications industry. To ensure the effective clustering of data samples with high dimensions, the raw features of the input data samples are encoded using the deep autoencoder network, and a more cluster-friendly representation is obtained. Then, clustering algorithms will be learned to segment customers, analyze each cluster, and give meaning to it, to provide a suitable strategy for analyzing customers' behavior. The experimental results on the dataset of the telecommunications company show that the customer segmentation method presented in this research has significantly higher efficiency (30% improvement of the Silhouette and 33% improvement of the Davies-Bouldin) and a more comprehensive clustering performance than the other traditional clustering methods.