Mining profitability of telecommunication customers and customer segmentation with novel data mining approach

Abstract

Telecommunication industry plays a vital role in the fast-moving modern world. At the same time, the industry is highly competitive because of multiple providers provide different solutions to their consumers. As a result, customers are rapidly moving from one service provider to another. Furthermore, human communications have been moving far from traditional calls and text messages to alternatives. Therefore, mobile operators are under real revenue threats as well as the risk of losing their potential customers. To solve this kind of issues, they need to increase their capabilities on understanding customer behaviour patterns and preferences, in order to achieve a high level of customer profitability and revenue. The major aim of this study is to cluster the customers based on profitability and develop a model to predict future customer’s profitability level and clustering the customers to provide different promotional packages. The current study is carried under three phases. The first phase is the comparison of different K-means algorithm and chooses the best one by using Within Cluster Sum of Square (WCSS) and processing time. The second phase is focusing on clustering the customers based on their behaviours by using K-means++ algorithm and develop the Artificial Neural Network (ANN) model to predict future customer’s profitability level. Finally, choose one of the early clustered customer group and apply K-means++ algorithm to provide different promotional packages. Dataset consists of 12,000 prepaid customer details with 15 different variables to cluster, train and test the model. Comparison of WCSS and process time, K-means++ is the best one for clustering. Confusion matrix used to evaluate the performance of ANN model and constructed model gives the accuracy of 97.3%. Existing researches use unsupervised or supervised learning algorithms separately. But this study integrates both algorithms and getting high accuracy result. Therefore, this model well fit for telecommunication industries.