Discovery of Usage Pattern from Mobile Call Data Using Clustering Approaches

Abstract

Mobile consumers’ expectations for data traffic continue to rise. Due to network components are finite, this presents a problem for the administration of the network's resources. Understanding the data consumption habits of mobile users can be an approach to resolving this issue. In this work, researchers collected a dataset, Call Details Record (CDR), which is free to download for the research purpose. The data collection is comprised of 17 attributes that are linked to 101,174 consumers and show whether that consumer converted. There is the total number of 8830 consumers. In the work, four clustering techniques are used to discover usage patterns of mobile call data from CDR dataset that are K-means clustering, fuzzy C-means clustering, divisive clustering and K-nearest neighbor (KNN). Monte Carlo validation methods are used to find optimal method. After choosing the optimum method in the previous stage, then this method allows access to the patterns that are predicted. Robustness of the suggested model is calculated at four parameters such as accuracy, precision value, recall and F-1 measure. Overall accuracy of the model is 89% which is better than previous compared model.