Global optimization applied to churn models

Abstract

This dissertation examines the probability that a subscriber churns from the current tariff he is subscribed to. These probabilities differ from one churn model to another and the optimal churn probabilities will be found by a global optimization algorithm and a standard optimization algorithm. When the optimal probabilities are obtained, a prediction of five or eight weeks is calculated, depending on the churn model. These predictions will then show which of the churn models implemented is the most accurate. In fact, the shifted-beta geometric (sBG) model is the most accurate and moreover, the global optimization algorithm performs better than the standard optimization algorithm. Modelling is done by use of Microsoft Excel and Matlab.