Churn detection via customer profile modelling

Customer profile modelling is an essential element of marketing in service applications to aid understanding customer behaviour, designing customized service plans, and preventing churn activities. In many service applications, profile data are often generated that consist of customer transactions over time. This paper proposes using a functional mixture model to profile customer behaviour in order to identify and capture churn activity patterns. A five-step procedure is proposed based on the functional mixture model, which includes: (1) standardizing profiles, (2) screening out uninteresting profiles, (3) projecting profiles onto a feature space represented by a set of basis functions, (4) applying clustering algorithms to the resultant coefficients in the feature space, and (5) identifying interesting profiles. It is shown that the proposed framework is effective for detecting churn activities in the telecommunication industry. The method can also be easily generalized to model sophisticated manufacturing processes for quality improvement.