Discovering usage patterns of telecommunication subscribers based on polytomous logistic regression

Abstract

Discovering patterns from telecommunication usage logs requires methodical procedures yet could reveal hidden information. In this paper, polytomous logistic regression has been applied to facilitate the discovery of these patterns. With an objective of determining characteristics of different categories of subscribers, data mining methods and techniques were used to extract and select data from collated text messages of network users. These were further normalized and cross-validated for model creation. Results disclosed hidden distinct attributes associated with each type of SMS users. These characteristics could facilitate the creation of new schemes of SMS packages as well as bundling of other services, hence a tipoff for improvement of business strategies for maximizing revenue of telecommunication companies.