| Abstract [eng] |
Customer relationship management is a comprehensive strategy to build, manage and strengthen long lasting relationships between business and its customers. One of the main tasks of this strategy is customer retention. The importance of this task is obvious. Firstly, loyal customers are considered to have the highest lifetime value because they generate the biggest portion of company‘s income. Secondly, acquiring new customers is much more expensive than retaining the loyal ones. Finally, it takes significant amount of time and financial resources to raise new customers’ loyalty and lifetime value. In this paper we consider a non-contractual customer relationship setting of a retail pharmacy. In such case, it is a complex task to define whether the customer has churned or not. In section 1 we perform a literature review to get acquainted with previous works regarding the customer loyalty and customer churn prediction. In section 2 we construct a customer churn analysis methodology starting with a raw customer transactional database. Generalizing customer churn definitions from the literature, we propose improved definitions of complete and partial churn, considering the behavioural characteristics of an individual customer. We also construct a set of behavioural characteristics including demographic, aggregated transactional, time series and time series discrete wavelet transformation coefficients. In section 3 we apply the proposed methodology on a retail pharmacy customer transactional database. We adapt logistic regression, support vector machines and random forest machine learning algorithms to build churn prediction models. We perform a stratified cross validation to compare prediction models of original and down sampled training sets, different sets of independent variables and different machine learning algorithms. Finally, we evaluate the predictive accuracy of performed prediction experiments and select the most efficient pharmacy customer churn prediction model. |
